Paper deep dive
YAQIN: Culturally Sensitive, Agentic AI for Mental Healthcare Support Among Muslim Women in the UK
Yasmin Zaraket, CĂ©line Mougenot
Intelligence
Status: succeeded | Model: google/gemini-3.1-flash-lite-preview | Prompt: intel-v1 | Confidence: 97%
Last extracted: 3/13/2026, 12:36:17 AM
Summary
YAQIN is a co-designed, agentic AI application developed to provide culturally and faith-sensitive mental healthcare support for Muslim women in the UK. By integrating Islamic psychology, Retrieval-Augmented Generation (RAG), and user-centered design, the tool addresses systemic gaps in mainstream mental health services, such as lack of cultural competence and therapeutic mistrust. The project involved a multi-stage design research methodology, including stakeholder interviews and co-design sessions, resulting in a chatbot and journaling tool that validates faith-based identities and improves therapeutic engagement.
Entities (5)
Relation Signals (4)
Yasmin Zaraket â developed â YAQIN
confidence 100% · This project investigates the design and evaluation of YAQIN... Yasmin Zaraket, Dr. Céline Mougenot.
YAQIN â supports â Muslim Women in the UK
confidence 100% · YAQIN is a co-designed AI-based application supporting culturally and faith-sensitive mental health engagement for Muslim women.
YAQIN â uses â Retrieval-Augmented Generation
confidence 100% · It incorporates a Retrieval-Augmented Generation (RAG) pipeline that connects Islamic knowledge.
YAQIN â integrates â Islamic Psychology
confidence 95% · It leverages AIâs anonymity... grounded in user-centred design principles where Islamic interpretations of psychology are key.
Cypher Suggestions (0)
No Cypher suggestions yet.
Abstract
Abstract:Mental healthcare services in the UK lack tools and resources to address the cultural needs of Muslim women, often leaving them feeling as though their values are pathologised and limiting trust and engagement [1]. Despite growing awareness of cultural competency, few interventions integrate Islamic frameworks into therapeutic support. This report investigates the design and evaluation of YAQIN, a co-designed AI-based application supporting culturally and faith-sensitive mental health engagement for Muslim women. With almost 1.9 million Muslim women in England in 2021, YAQIN responds to a gap in care [2]. It leverages AIĆ anonymity and continuous support through a faith-aware chatbot and guided journaling tool grounded in user-centred design and Islamic psychology. The YAQIN design research methodology comprised three stages: contextual investigation and literature review, user research with N=14 stakeholders including Muslim women and mental health experts, and prototype development informed by deductive thematic analysis, personas, journey maps, and design specifications. Evaluation involved a co-designed user study with five participants: four Muslim women and one mental health expert who reviewed therapeutic alignment and cultural sensitivity after using the chatbot prototype. Feedback focused on tone, faith relevance, emotional resonance, and the Retrieval-Augmented Generation pipeline enabling contextual continuity. Participants highlighted YAQINĆ ability to bridge cultural gaps in trust and therapeutic confidence. Feedback included suggestions of including linguistic diversity and routine-based guidance. This project demonstrates how culturally sensitive AI can improve mental healthcare accessibility and trust for marginalised communities and highlights the potential of faith-integrated technology in healthcare innovation.
Tags
Links
- Source: https://arxiv.org/abs/2603.07709v1
- Canonical: https://arxiv.org/abs/2603.07709v1
Trouble viewing inline? Open PDF directly â
Full Text
79,743 characters extracted from source content.
Expand or collapse full text
March 2026 YAQIN: Culturally Sensitive, Agentic AI for Mental Healthcare Support Among Muslim Women in the UK Yasmin Zaraket, Dr. CĂ©line Mougenot 1 Dyson School of Design Engineering, Imperial College London Abstract Mental healthcare services in the UK lack the tools and the resources to meet nuanced cultural needs of Muslim women to avoid them feeling their values are pathologised and, consequently, miss opportunities of trust and engagement [ 1]. Despite growing awareness of cultural competency, few interventions meaningfully integrate Islamic frameworks into therapeutic support. This report investigates the design and evaluation of YAQIN, a co-designed AI-based application that supports cultural and faith-sensitive mental health engagement for Muslim women. With the population of Muslim women in England increasing annually from almost 1.9million in 2021, YAQIN responds to a significant gap in care [ 2]. It leverages AIâs anonymity, continuous support, and contextual intelligence to offer a faith- aware chatbot and guided journaling tool, both grounded in user-centred design principles where Islamic interpretations of psychology are key. The YAQIN Design Research Methodology encompasses three iterative stages. The first was rigorous contextual investigation and review of available literature. Second, initial user research included semi-structured interviews with N=14stake- holders, comprising both Muslim women who had sought mental health services and mental health experts. YAQIN analyses the insights through deductive the- matic analysis and translated them into personas, journey maps, and design spec- ifications which informed its first prototypes. This project conducted the evaluation through a co-designed user study involv- ing five participants: four Muslim women who engaged in a live interaction and co-design session, and one mental health expert who provided targeted critique on therapeutic alignment and cultural sensitivity after using the chatbot proto- type. YAQIN assessed the design of the chatbot based on tone, faith relevance, and emotional resonance based on key co-design feedback focused on personalised reflection prompts, Quranic contextualisation, and the Retrieval-Augmented Gen- eration pipeline used for contextual continuity. Participants praised its ability to bridge cultural gaps, particularly in trust, articulation, and therapeutic confidence. YAQIN successfully fosters emotional reflection, validates faith-based identity, and supports therapy readiness as ev- idenced by the evaluation findings. Moreover, feedback highlighted areas for growth, such as supporting linguistic diversity and offering deeper routine-based Second author:CĂ©line Mougenot, Senior Lecturer, Associate Professor, Imperial College London1 arXiv:2603.07709v1 [cs.HC] 8 Mar 2026 2YAQIN guidance. YAQINâs design and development implemented this feedback into the final iteration. This project leverages the value of culturally-sensitive AI to improve mental healthcare accessibility and trust for marginalised communities. It contributes to current systems by operationalising Islamic psychological conceptual in digital care tools, highlighting the broader potential of contextual faith-integrated, user- informed technology in healthcare innovation. Keywords:Mental Health; Muslim Women; Cultural Competence; AI; RAG; Islamic Psychology; Empathic Design; User-Centred Design; Therapeutic Trust 1Introduction Mental healthcare resources in the UK are strained and barely meet basic demand, leaving little room for nuanced services and cultural accommodations. Challenges such as systemic bias, cultural stigma, and the lack of religiously informed therapeu- tic models contribute to potential disengagement from mainstream services [1,3]. Despite increased recognition of cultural competency in healthcare, many services lack the training, time, and resources needed to meaningfully accommodate faith- sensitive perspectives [ 4]. This project addresses the disconnect and focuses on bridging the gap between conventional therapy andthe needs of Muslim womenby exploring how recent ad- vances in artificial intelligence (AI) and innovation can support mental healthcare interventions with cultural sensitivity. In2021â22, only2.6% of almost3.8million Muslims in England, with over1.9million women included, completed their referred mental healthcare plans [ 5,2], highlighting a critical need for more inclusive and faith-aligned pathways to mental healthcare. This project develops and utilisesnovelculturally and faith-sensitiveagentic AI to bridge the disconnect between Muslim women and mainstream mental healthcare, specifically focusing on client user experience (UX). While initiatives like Muslim Youth Helpline [ 6], Inspirited Minds [7], and the Muslim Womenâs Network UK [8], offer valuable community support, they are often limited to set hours and con- strained by volunteer capacity [ 6,7]. YAQINâs use of AI complements these efforts by offering scalable, on-demand, and personalised support. Termed YAQIN,ÙÚŻà¶à”čArabic for certainty, it incorporates a Retrieval-Augmented Generation (RAG) pipeline that connects Islamic knowledge, Islamic interpretations of psychology, and novel journaling features within a chatbot agentic interface. This ensures contextual continuity and dynamically surfaces relevant and culturally as- suring knowledge during client interactions. Motivation:Consider Lina, a student struggling with exam anxiety, unable to explain to her therapist how supplication helps her feel calmer and less anxious. YAQIN offered a bridge by reframing supplications as breathing-based recitation exercises with meaningful reassuring phrases. This improved communication and engagement with her therapist. The central research question is: How might we co-design a culturally and faith-sensitive AI tool, informed by Yasmin Zaraket3 Islamic resources, to support Muslim women navigating mental healthcare in the UK? YAQIN targets cultural minority groups and focuses on UK-based Muslim women as a first success story. It is an innovative, culturally rooted, AI-powered application built to validate and support usersâ cultural, emotional and spiritual needs. Its RAG presents a contextualisation framework to include cultural sensitivity and leverages Islamic interpretations of psychology to bridge an observable gap in mental health- care for Muslim women. This framework is adaptable to other similar contexts and service gaps. The YAQIN Design Research Methodology (DRM) framework combines litera- ture review, semi-structured interviews with seven Muslim women and seven mental healthcare experts, and deductive thematic analysis in the Descriptive Study I stage, concept and initial prototype development in the Prescriptive Study stage, and eval- uation, iterative prototyping, and discussion reflecting on broader implications and future potential in the Descriptive Study I stage. They appear in Sections3-8, respectively. Insights from each stage translate into user personas, journey maps, and design specifications, informing the development of YAQIN. In the Descriptive Study I stage, four participants joined a co-design session and one expert participated in usability testing to assess the YAQIN chatbot therapeutic relevance, emotional res- onance, and usability. In summary, YAQIN, implemented, successfully deployed, and tested on Hugging- face, with a click-through UI, responds to a pressing need for inclusive, empathetic, and contextually grounded mental health solutions. By leveraging faith-sensitive psychological paradigms and user-centred design, YAQIN revolutionises culturally- sensitive AI tools in therapeutic engagement. It contributes to a broader conversa- tion on how healthcare technology can be ethically and critically aligned with the lived realities of marginalised communities. Figure 1: Motivating Use Case with Lina Persona 4YAQIN 2Project Aim & Objectives This project aims to explore how AI can support culturally and faith-sensitive men- tal health care. By focusing on Muslim womenâs experiences, it develops an ethi- cal, spiritually grounded agentic intervention that enhances therapeutic trust and emotional validation through contextually relevant, Islamically informed design and interaction. Objectives: âąUnderstand emotional and cultural barriers Muslim women face accessing men- tal healthcare. âąIdentify themes, needs, and aspirations on the relevance, trust, and emotional experiences of Muslim women through in-depth stakeholder interviews. âąDevelop user personas and journey maps grounded in interview qualitative data synthesis via deductive thematic analysis. âąMap the pain points and highlighted insights to opportunities in leveraging emerging technology like Large Language Models (LLM) and agentic AI inte- grations. âąDesign and prototype YAQIN, integrating agentic AI to address identified needs. âąEvaluate YAQIN effectiveness in fostering cultural sensitivity and competence in therapeutic contexts. 3Literature Review Cultural competency is an emerging concept in mental healthcare. Statutory ser- vices often feel âmonoculturalâ and lacking religious competence, particularly for ethnic minority communities who do not see their values or worldviews reflected in mainstream provision [ 9]. These concerns are echoed by both the NHS Race and Health Observatory (RHO) and the British Medical Association (BMA), who note that mental health services remain under-resourced in addressing the needs of ethni- cally diverse communities [3,5]. As a result, many individuals from ethnic minority backgrounds avoid care due to mistrust and fear of racist or culturally uninformed treatment [ 3,5]. The Muslim Mind Collaborative (MMC) has argued that the âneed for culturally and faith-sensitiveâ mental healthcare must shift from a specialist concern to a mainstream clinical priority [ 10]. Recent data illustrate the impact of this gap: in 2021â2022, only 2.6% of Muslim clients completed therapy plans, compared to 18.4% of Christians and 38.9% of non-religious clients [5]. Focusing on UK Muslims, 44% of them felt faith-related issues were inadequately addressed in mainstream counselling [ 5]. A key factor is lack of religious inclusivity. Ahmad et al. argue that integrating Islamic perspectives can bridge important gaps in therapy [11] and improve mental healthcare outcomes [12]. Reviews of cultural competence in mental healthcare warn that static, checklist- based models often fail to capture the fluid, intersectional realities of service users and may unintentionally realise cultural stereotypes [13]. These concerns support arguments that cultural responsivity in technology-enabled mental health systems Yasmin Zaraket5 must move beyond generic notions of âcultural competenceâ toward ongoing, context- aware adaptation that recognises usersâ identities, histories, and power relations [14]. Narrative reviews of AI for positive mental health further highlight that, while AI systems can improve access, monitoring, and personalisation, these benefits will not be equitably distributed unless cultural and contextual factors are explicitly embedded into system design and evaluation [15]. 3.1Cultural Inaccessibility and Systemic Barriers for Ethnic Minority Women The Race Equality Foundation (REF) reports that women from diverse ethnic back- grounds are âdisproportionately affected by adverse mental health outcomes and underutilise servicesâ due to a combination of systemic discrimination, cultural in- accessibility, and mistrust [1]. Muslim women in particular may feel alienated in therapy settings when their cultural norms and values are pathologised, minimised, or dismissed [9]. Fears around judgement, stigma, shame, and breaches of confiden- tiality further inhibit engagement with mental healthcare services [10,12]. Recent work in AI mental health design highlights how these barriers can be amplified or alleviated trrough technology. The SMILE platform, for example, illus- trates how an AI-driven mental health system can embed multilingual support and culturally sensitive therapeutic tools to reach diverse populations, explicitly posi- tioning cultural adaptation as the centre of design rather than an optional layer [ 16]. Cross-cultural validation studies of AI-based mental health assessments show that linguistic nuances, culturally specific symptom expressions, and biased training data can lead to systematic misdiagnosis or under-recognition of distress among minori- tised users [ 17]. Building on these concerns, overviews of AI in mental health warn that unless gender, race, migration status, and faith-based experiences are deliber- ately represented in datasets and outcome measures, AI systems risk reinforcing the existing invisibility of women and ethnic minorities in both access to and the quality of mental healthcare [ 15]. Identified themes:Confidentiality and Trust; Changes in Access and Acceptance 3.2Breaking Trust: Bias, Misinterpretation, and Cultural Disconnect Trust is particularly diïŹicult to establish where Muslim womenâs experiences are shaped by culturally uninformed care and provider bias [18]. Studies with Iraqi Muslim women in clinical settings report frequent experiences of being misunder- stood, especially when norms around modesty and communication are overlooked or challenged by clinicians [ 18]. Healthcare staff sometimes question Muslim womenâs credibility, reinforcing feelings of marginalisation in clinical spaces [19]. Muslim per- spectives on mental illness typically integrate spiritual, social, and clinical dimen- sions, whereas dominant clinical frameworks often adopt secular biomedical models that foreground pathology over lived meaning [ 20]. Muslim mental health practi- tioners therefore advocate embedding cultural and religious understanding within NHS care practices [21]. Among Muslim minorities, stigma and a lack of cultural competence within ser- vices contribute to the underutilisation of mental healthcare [22]. This can worsen 6YAQIN academic and personal wellbeing, particularly for university students who may ex- perience heightened cultural dissonance and fear of judgement [23]. Many Muslim women fear that disclosing personal struggles could harm family reputation and social standing, discouraging them from seeking help or being open in counselling sessions [ 24]. In this context, building therapeutic trust requires cultural compe- tence, empathetic engagement, and respectful professional relationships that trans- form clinical settings from spaces of scrutiny into âsafe circles of healing.â These issues are mirrored in debates about AI in mental health. Ugar and Malele argue that there should not be a homogeneous or generic design for AI and ma- chine learning systems diagnosing mental health disorders, because such systems embed value-laden judgements about what constitutes disorder that differ across cultures [25]. They show that mental health diagnosis depends on both âfactual evi- denceâ and culturally situated value judgements, so AI systems trained primarily on Western notions of pathology may misclassify culturally normative experiences and undermine trust in regions such as sub-Saharan Africa [25]. Broader work on cultural responsivity in technology-enabled services similarly warns that generic cultural com- petence frameworks can obscure within-group differences, and that trust depends on acknowledging usersâ specific histories, identities, and structural marginalisation [14]. Extending this insight, early research on âcultural promptingâ for AI mental health agents shows that explicitly instructing large language models to attend to usersâ cul- tural backgrounds can enhance perceived empathy and appropriateness of responses, offering a tangible mechanism for rebuilding trust in AI-mediated support [ 26]. Identified themes:Therapist and Client Prejudice; Stigma and Cultural Taboo; Creating a Safe Environment 3.3Faith-Aware Therapeutic Adaptations and Inclusive Clinical Practice Improving cultural competence is associated with stronger therapeutic relation- ships and better outcomes for Muslim clients [ 27]. Rahman et al. emphasise the importance of âtruly culturally sensitive care,â advocating for faith-informed, client-centred approaches that treat religious identity as a core part of the person rather than an optional add-on [ 28]. Within many Muslim communities, stigma around âfamily shameâ complicates disclosure in therapeutic contexts, reinforcing concerns about confidentiality and moral judgement [ 24]. Suggested adaptations include gender-concordant therapists and space for spiritual discussion when appro- priate; Muslim clients often describe such arrangements as âreassuringâ and âtrust- buildingâ [29]. Spreitzerâs psychological empowerment framework highlights meaning, compe- tence, autonomy, and impact as key dimensions across cultural contexts [ 30]. This framework reinforces the need to validate clientsâ identities and provide choices that reflect their values, and it is particularly relevant for designing inclusive and respon- sive mental healthcare that mitigates religious and ethnic disparities. Therapists also need to recognise that distress may be expressed through cultural or religious idioms-such as religious obsessive-compulsive disorder (OCD) or moral struggle [ 4]. Language sensitivity, religious literacy, and familiarity with Islamic worldviews be- come critical for avoiding ruptures in the therapeutic relationship. Yasmin Zaraket7 There are parallel developments in AI-mediated training and simulation that show how cultural competence can be supported through technology. Co-design work on a chatbot for culturally competent clinical communication demonstrates that AI can simulate diverse patients and provide structured feedback guided by cultural com- petence models, helping trainees reflect on empathy, listening, and cross-cultural understanding, while also revealing current limitations such as the lack of non-verbal cues and overly agreeable virtual patients [31]. Behavioural health projects using AI and machine learning as deliberate-practice tools aim to improve cliniciansâ cultural competence by giving targeted feedback on language, micro-aggressions, and respon- siveness; here AI is positioned as a trainer helping them recognise and correct bias rather than a neutral replacement for therapists [32]. Policy and ethics commentary on culturally appropriate care in the age of AI stresses that culture is flexible and evolving, and, therefore, AI systems and training tools must be designed to adapt to changing identities and community norms rather than freezing static stereotypes into algorithms [33]. Identified themes:Improvement in Inclusivity and Representation; Language and Communication Barrier; Client Preferences; Conventional Models vs. Cultural Re- alities; Training, Supervision, and Peer Networks 3.3.1Integrating Islamic Epistemologies Into Modern Mental Health- care To adequately integrate religious and culturally sensitive care into therapy and coun- selling, it is essential to engage with Islamic interpretations of psychology and mental wellbeing. Islamic psychology emphasises thenafs(psyche) as eternal and spiritu- ally anchored, contrasting with secular models that focus primarily on the mind as a cognitive or biochemical entity [ 4]. Muslim clients often interpret mental distress through a sense of divine purpose, shaping their expectations about the role and limits of therapy [ 28]. Culturally adapted CBT models illustrate how faith-based interpretations can be meaningfully aligned with evidence-based psychological prac- tice [34]. The concept oftazkiyat al-nafs(self-purification) encourages aligning the inner self with divine guidance to attain emotional and spiritual balance [4]. Historical figures such as Abu Zayd al-Balkhi, a 9th-century scholar, distinguished between physical and psychological illness and proposed proto-cognitive therapeutic approaches, with significant convergence between his diagnostic criteria and contemporary frameworks like the DSM [4]. He emphasised how thoughts influence mood and advocated preventative emotional care, making his work one of the earliest documented efforts to integrate psychotherapeutic insight within religious and medical frameworks [ 4]. Contemporary Islamic mental healthcare frameworks build on these concepts. Quranic teachings and religious coping practices-such as prayer (salat), remembrance (dhikr), and supplication (dua)-have been shown to reduce stress, anxiety, and de- pressive symptoms when integrated appropriately [ 35,36]. Faith-sensitive assess- ment tools aligned with Islamic value systems support more accurate and culturally meaningful evaluation of mental states [4]. However, there remains a shortage of culturally competent practitioners and limited time and resources in mainstream 8YAQIN therapy [3]. These constraints highlight the need for scalable interventions, such as AI-assisted, faith-informed tools, that can augment both clients and practitioners in delivering culturally sensitive mental healthcare [37]. When AI systems interact with religious content, additional risks arise. AI may misinterpret scripture or generate unauthoritative or misleading spiritual guidance, raising serious concerns among religious communities [38]. There is particular worry that AI tools could dilute or bypass traditional sources of religious authority if not carefully constrained [38]. Ethical discussions emphasise a strong requirement for AI to protect user privacy when handling spiritually sensitive disclosures, ritual practices, or matters of belief [38]. When appropriately designed, AI can expand access to religious knowledge without substituting human scholars, but this requires explicit ethical guidelines to ensure authenticity, accuracy, and faith-aligned bound- aries [38]. Identified themes:Redefining Psychological Concepts with Islamic Terminology; Prophetic Stories and Quranic Anchors 3.3.2Faith-Sensitive Design in Mental Healthcare Faith-sensitive approaches can help bridge critical gaps between Muslim clients and conventional mental healthcare models. Islamic frameworks centred on thenafs (self),qalb(heart), andâaql(intellect) offer spiritually coherent models for under- standing and treating emotional distress [ 4]. When metaphysical constructs such astawakkul(trust in God) andsabr(patience) are explicitly integrated into care, Muslim women often report increased trust in the therapeutic process and greater acceptance of interventions [ 4]. Quran-based interventions such as structured recitation, guided reflection, and spiritual mentoring can support mental wellbeing, particularly when coupled with reflective journaling practices that help individuals clarify emotions and internal conflicts [ 39,36]. Therapy becomes more effective when Islamic narratives and literature are embedded within clinical interactions, shifting the model to serve the clientâs worldview rather than forcing clients to bracket their faith. In turn, this builds therapist-client relationships that respect both clinical needs and faith-based identities. Identified themes:Integration of Religious Practices in Sessions; Faith-Sensitive Client-Centred Practice 3.4The Role of AI in Expanding Culturally Sensitive Mental Healthcare To address gaps in cultural sensitivity and accessibility, this project examines how AI can support the mental healthcare needs of underserved cultural groups-particularly Muslim women-by embedding cultural and religious awareness into digital interven- tion design. Yasmin Zaraket9 3.4.1Opportunities for Faith- and Culture-Aware AI Interventions Recent advances in multilingual conversational AI have opened access to nuanced cross-cultural knowledge that was previously diïŹicult to operationalise at scale. The UK Parliamentary OïŹice of Science and Technology notes that âAI-based tools could support healthcare professionals to provide personalised care more quickly,â partic- ularly for people underserved by traditional services [40]. Ahmad et al. argue that a well-designed AI tool in mental healthcare can âreduce systemic delay, ensure anonymity, and increase therapeutic accessâ for underserved Muslim women in the UK [37]. Several interventions showcase how culturally sensitive design can be realised in practice. The Chatbot-Assisted Self-Assessment (CASA) intervention, co-designed with ethnic minority communities, enables users to self-assess mental wellbeing and access tailored medical advice [41]. Tunggala proposes an Islamic ethics-based frame- work for aligning AI communication with religious values, offering a way to balance innovation with cultural and theological sensitivity [42]. In the broader app ecosys- tem, Calm Health, Headspace, and Wysa provide accessible and well-designed AI or wellness tools, but largely lack faith- and culture-sensitive framing [43,44,45]. Faith- informed platforms such as Muslim Moodfit and Ruhi integrate Islamic content, yet do not offer robust personalisation or co-designed features [ 46,47]. Inspirited Minds offers culturally aware therapy but does not currently include AI-based self-help or reflective tools [ 48]. These gaps collectively motivate YAQINâs focus on combining AI with co-designed, culturally and faith-sensitive support for Muslim women. HCI work on tracking during Ramadan provides further insight into design needs. Studies of Muslim womenâs menstrual and religious tracking behaviours show fre- quent value conflicts between religious obligations (such as fasting) and physical or hormonal health needs, leading many users to âhack togetherâ personal tracking workflows to align mainstream apps with fiqh-based rules, highlighting significant un- met design needs, as existing tools fail to represent nuanced religious and menstrual factors such as fasting exemptions or prayer interruptions, often resulting in stress, guilt, and self-judgment [ 49]. At the interactional level, conversational agents have been shown to increase womenâs participation and articulation of ideas in mixed- gender online discussions, suggesting that chatbots can amplify under-represented voices and mitigate social barriers tied to gender [50]. Identified themes:Brainstorming Features 3.4.2Ethical, Cultural, and Technical Challenges Despite its promise, AI in mental healthcare introduces significant ethical and cul- tural challenges. Emke warns that cultural insensitivity in translation algorithms may reinforce, rather than remove, linguistic and contextual errors, âleading to a de- terioration in output qualityâ [ 51]. AI models trained primarily on Western datasets may not generalise to culturally diverse users [52]. Data biases, privacy risks, and misinterpretation of user intent can render AI interventions exclusionary and harm- ful for vulnerable groups [ 53]. YAQIN attempts to mitigate these issues by using co-design methods and restricting its responses to a curated semantic database, 10YAQIN enabling retrieval-augmented generation (RAG) that is more culturally and theolog- ically aligned [54]. Research on culturally aware AI design shows that systems often fail when they do not meaningfully embed the cultural norms, communication styles, and value systems of their intended users [55]. This gap leads to biased data use, misinterpre- tation of cultural idioms, and outputs that can marginalise minority communities. To avoid this, culturally aware AI must be developed through participatory design with the target community, ensuring the system reflects their lived experiences [55]. Effective localisation requires more than small interface changes; it involves adapt- ing tone, language, examples, and reasoning patterns to the relevant cultural and religious context. Ongoing community evaluation is also essential to maintain value alignment and prevent cultural erasure, ensuring cultural context shapes the modelâs behaviour rather than only its surface features [ 55]. Identified themes:Communication and Language; Data and Privacy 3.5Summary and Research Question Across this literature, there is clear evidence of persistent cultural, systemic, and religious misalignments between Muslim women and mainstream mental healthcare. Barriers include institutional bias, stigma, fear of judgement, language challenges, and the absence of faith-integrated care. At the same time, both clinical and tech- nological literatures highlight the risks of culturally naive AI systems and the im- portance of participatory, context-aware, and faith-sensitive approaches. With limited resources and structural inequalities, bridging these gaps requires both culturally informed therapeutic models and scalable innovations that address access, representation, and inclusivity at every stage of mental healthcare delivery. This motivates the central research question of this project: How might we co-design a faith- and culturally-sensitive AI tool, informed by Islamic resources and personalised feedback to improve the relevance, trust, and emotional experiences of Muslim women navigating mental healthcare in the UK? 4Methodological Approach This paperâs methodological approach leverages the DRM [56], which provides a structured yet flexible framework tailored for iterative research and the development of evidence-based design solutions. DRM, illustrated in Figure2, is characterised by three interconnected stages: Descriptive Study I, Prescriptive Study, and Descriptive Study I. Each stage strategically builds upon insights from the preceding stage and allows for refinement and robust validation of outcomes. 5User Research & Insights This section presents primary user research following a literature review on cultural and religious factors in mental healthcare. Yasmin Zaraket11 Figure 2: DRM Outline for YAQIN 5.1Research Methodology This user research explores Muslim womenâs experiences navigating UK mental healthcare systems, aiming to enhance the relevance, trust, and emotional support. User interviews offer a powerful means of capturing nuanced, context-rich insights that cannot be accessed through literature alone [57]. The interviews aim to: 1.Deepen and validate insights from the literature by exploring the lived expe- riences of stakeholders. 2.Identify key barriers Muslim women encounter when seeking mental healthcare and how trust can be built to overcome them. 3.Understand the role of Islamic teachings and concepts in mental health expe- riences, and how these can be effectively integrated into therapy. 4.Gather specific recommendations for designing culturally sensitive mental health interventions tailored for Muslim women. 5.1.1Interview Structure & Ethics This project conducted each30â45minute interview in person or online over the course of three months, enabling a systematic exploration of culturally sensitive mental healthcare experiences. The interviews focused on two primary stakeholder groups with the following inclusion criteria: 1.Muslim women in the UK (Clients) with experience accessing or consideration of mental healthcare services 2.Mental healthcare psychotherapists or researchers (Experts) This project recruited interviewees via community networks and university events. This may introduce sampling bias but it is important for this project to have this niche group of interviewees. The average age for experts was39.4years (SD =12.0), and for clients,21.7years (SD =1.1). 12YAQIN Table 1: List of Interviewees and Their Occupation Muslim Women Clients C1C2C3C4C5C6C7 Under- graduate TeacherMedical Student Under- graduate Masterâs Student Under- graduate Under- graduate Mental Health- care Experts E1E2E3E4E5E6E7 Psychologist, Psy- chothera- pist Pscho- therapist, Counsellor Medical Re- searcher CounsellorClinical Team Lead, Pscyho- therapist Psychology Re- searcher, Islamic Scholar Mental Health Pharma- cist Data saturation in homogenous groups is typically achieved with around7inter- views [58]. This project conducted N=14interviews with seven Muslim women and seven experts, ensuring thematic depth without oversampling. To ensure wellbeing protection, interviewees were excluded if mental health discussions posed a risk of distress or emotional harm. Table1lists the participating interviewees and their roles. The Imperial College Research Ethics Committee approved the ethics application prior to interviews. This project did not actively ask for personal mental health disclosure to minimise emotional risk or triggering. Such disclosures would have required separate approval from the Science, Engineering and Technology Research Ethics Committee (SETREC). The interviews began with verbal informed consent to record the session, knowing that they could withdraw at any time. 5.1.2Deductive Thematic Analysis Approach Following Braun and Clarkeâs six-phase model of familiarisation, coding, theme development, review, definition, and reporting, the deductive thematic analysis method supports a systematic yet flexible exploration of meaning in qualitative data [ 59]. This project identifies key themes from Section3like stigma, cultural barriers, trust in therapy, and integration of Islamic teachings that directly relate to the cod- ing phase in deductive thematic analysis. Figure 2below shows the codes grouped into broader thematic clusters. This structure grounded the interview analysis and validated literature insights through real-life narratives. 5.2Findings This project reveals consistent challenges and opportunities in culturally sensitive mental healthcare for Muslim women in the UK through deductive thematic analysis of user interviews. By engaging clients and experts, this analysis captures how cul- tural and religious values shape therapeutic engagement, expectations, and overall outcomes. The tension between culturally rooted stigma and the need for non-judgmental and accessible mental health support was the most consistent pain point. As one expert noted, Yasmin Zaraket13 Table 2: Thematic Analysis Code and Theme Breakdown from Literature Review ThemesCodes Identified in Liter- ature Review Changes in Cultural and Religious Diversity Acceptance Improvement in Inclusivity and Representation Changes in Access and Accep- tance Stigma, Judgement, and Barriers to Care Stigma and Cultural Taboo Therapist and Client Preju- dice Islamic Knowledge in Therapy Integration of Religious Teachings in Sessions Prophetic Stories and Quranic Anchors Faith-InclusiveClient- Centred Practice Current Clinical Practices and Framework Clashes Conventional Models vs Cul- tural Realities Training, Supervision, and Peer Networks Trust and Safety in Therapy Sessions Confidentiality and Trust Creating a Safe Environment Communication Barrier in Therapeutic Relationship Language and Communica- tion Barrier Redefining Psychological Concepts with Islamic Termi- nology Client Preferences Design Considerations for Mental Health Tools Communication and Lan- guage Brainstorming Features Data and Privacy âHistorically, ethnic minorities do not reach out for mental health be- cause itâs not part of their kind of thinking [...] support is usually done within the community or spiritual communityâ (E5) This validates the literature describing mental healthcare stigma as particularly pro- nounced in collectivist and religious communities where struggles may be dismissed as signs of weak faith or poor character. Interviewees also highlighted the barriers to finding appropriate therapy. They expressed frustration with systems that felt unintuitive or culturally misaligned with barriers like cost, availability, and discomfort navigating conventional clinical norms. One client spoke of her struggle seeking help through the most affordable, accessible pathway; her university counselling service. 14YAQIN âThe system wasnât really easy to access [...] there wasnât a clear place on the website to go toâ (C3) While C3 ended up seeing a non-Muslim counsellor, she and the majority of interviewed clients expressed a preference for seeing one thatâs Muslim. As C4 explained: âYou can much more easily share [your struggles] with someone else who obviously grew up knowing about Islam or your culture at the same time as youâ (C4) Both experts and clients noted that therapists of any background could offer effective care so long as they approached clientsâ faith as meaningful and valid. This supports literature advocating for âcultural humilityâ over âcultural matchingâ. Effective therapy relies on client-centricity. Cultural mismatch was also evident when clients felt pressured to adopt therapeu- tic recommendations that conflicted with religious or cultural norms. As one client recalled: âLetâs say youâre talking about how youâre going to put boundaries with [...] your mum. The [therapist] would suggest: âleave the houseâ, âgo move outâ. Thatâs not an option that we want to consider, not only because [...] itâs not Islamically the right thing to do. It doesnât make sense.â (C2) This underscores the need for therapists to frame behavioural strategies within the bounds of clientsâ ethical and familial frameworks rather than imposing normative values of individualism or autonomy. Clients appreciated therapists who connected Islamic practices with psychologi- cal tools. Both clients and experts emphasised the power of integrating concepts likedua(supplication),tawakkul(leaving matters in Godâs hands), and Prophet Muhammadâs emotional experiences. One expert remarked, âEven the Prophet suffered with sadness [...] thereâs nothing wrong with thatâ (E7), reframing emotional struggles as spiritually significant rather than shameful. Another expert invoked a Quranic verse to highlight the therapeutic potential of patience, saying, âTake the verse âFaáčŁbir áčŁabran jamÄ«lanâ [So endure patiently with a con- tent patience (Quran 70:5)]. It acknowledges hardship yet commands [Muslims] to respond withsabr[patience]. Recognisingsabras a thera- peutic skill reframes suffering as an opportunity for growth.â (E6) One expert shared how integrating religious values supported a client with long- term depression. â[My] client described qualities she wished to cultivate likeáž„ilm(forbear- ance and abstinence), and I introduced ahadith[(a narration)] from Imam Jafar al-Sadiq on the seven aspects ofakhlaq(personality ethics), linking spiritual ideals to therapeutic goals.â (E1) Yasmin Zaraket15 Figure 3: Personas (top) MP1 Maryam: Muslim Woman Attending Therapy Sessions with a Muslim Therapist After a Non-Muslim Therapist Experience and (bottom) MP2 Lina: Muslim Woman Struggling to Access Mental Healthcare [ 63] 5.2.1Persona & User Journey Mapping The analysis uses personas and user journey maps, core user-centred design tools, to translate interviews into design specifications. Personas translate lived experiences into relatable profiles that support empathy-driven design [ 60], while journey maps reveal emotional, cultural, and systemic barriers over time [61]. These methods align stakeholder needs with culturally-sensitive service design [ 62]. Insights from thematic analysis and contextual research informed the four per- sonas in Figures3and4and their simplified journey maps in Figures5and6. 5.3Discussion The user journeys identify core pain points at every stage such as diïŹiculty accessing culturally resonant care, fear of judgment, limited integration of faith, and emotional disconnects. MP2âs reluctance to continue therapy after feeling misunderstood or TP2âs hesitation in engaging with unfamiliar religious language underscore a broader misalignment between current mental healthcare models and the realities of Muslim women seeking mental healthcare. These pain points echo literature findings that the Muslim womanâs experience in therapy is at risk of being culturally dissonant, inaccessible, or emotionally un- safe [ 60,61]. For instance, there is a large increase in clientsâ preference for therapists who inherently understand their values and struggles. This accentuates the design need for safe ways to express faith (likeduaor Quranic reflection) without risking being pathologised. 16YAQIN Figure 4: Personas (top) TP1 Layla: Muslim Woman Therapist Working with Diverse Clients and (bottom) TP2 Alex: Non-Muslim Therapist Supporting Faith-Inclusive Practice [63] Figure 5: MP1 Maryam and MP2 Lina User Journey Maps: Different Journeys of Navigating Starting or Restarting Therapy Sessions Figure 6: TP1 Layla and TP2 Alex User Journey Maps: Different Journeys of Navigating Clients with Familiar or Unfamiliar Cultural Backgrounds Issues such as emotional withdrawal during early therapy experiences, as seen in MP1âs journey, point to a need for clear, culturally and religiously attuned communi- Yasmin Zaraket17 cation. Likewise, the desire for continued engagement outside therapy sessions backs the potential for structured, therapist-linked support tools that enhance therapy and do not replace it. 5.3.1Summary In summary, the user journeys and personas form the foundation for designing an intervention that reduces cultural friction, integrates meaningful Islamic constructs, and supports both emotional and therapeutic continuity. Each step in the journey should help shape YAQIN to align with the values, vulnerabilities, and healing aspirations of its users. 5.3.2Limitations Although the sample size of14participants satisfies the saturation criteria, this project recognises that the findings may lack generalisation to Muslim populations outside the UK. While interviewees matched the defined target user-group, limited geographic and ethnic diversity narrowed the breadth of perspectives. This is a valuable consideration for future developments and broader application. However, the focused scope enabled deep engagement with the target users, and their varied experiences offer strong directions for designing culturally sensitive interventions within the context of this project. 6YAQIN App Design & Development This section details the features of YAQIN and the interacting components that realise them. They fulfill design specifications and embody target user values, con- cerns, and aspirations. They translate cultural sensitivity into user-centred app experiences. 6.1Design Specifications & Success Criteria YAQIN must satisfy the design specifications and their associated constraints to function effectively. The descriptive study of Section5results in these character- istics and constraints through the synthesis of user interviews, cultural analysis, and psychological literature. The resulting12specifications show in Figure7with their rationales. They include integrating culturally relevant paradigms and Islamic teachings, supporting cognitive behavioural strategies, and preserving data privacy. 18YAQIN Figure 7: Design Specifications for YAQIN Intervention 6.2Ideation & Concept Generation Participants frequently described misfits between Muslim moral and relational values and assumptions embedded in conventional mental healthcare practice. For example, Yasmin Zaraket19 advice to cut ties with family by therapists, is often religiously unacceptable, and culturally incompetent. One user summarised this discord: âIt doesnât make sense [...] itâs not Islamically the right thing to doâ (C2). These misfits inspiredYAQIN, an AI-agentic application that bridges emotional and spiritual-cultural needs in a resource-strained mental healthcare system. In this prescriptive study, the resulting looks-like and works-like prototypes focus on simulating emotionally diïŹicult conversations in a faith-sensitive, safe digital space. 6.2.1Branding & Visual Identity YAQIN(ÙÚŻà¶à”č) is an Arabic term that connotes a deep, unwavering sense of spiritual certainty. In epistemology,yaqinreflects a tranquil and resolute state of cognition achieved through understanding oneself. The name reflects YAQINâs goals to sup- port users through emotional confusion and align with their values. It implies clarity and healing can coexist with struggle and introspection. The visual language of YAQIN as illustrated in Figure8complements this epis- temic grounding. A muted palette fosters psychological calm and reduces cognitive load on the interface [ 64]. Design specifications met:1, 2, and 8. 6.3Main Feature Specification The following subsections describe the features that operationalise YAQINâs objec- tive: to help clients better understand and articulate their emotional and spiritual experiences. Features leverage culturally-sensitive reasoning, Islamic concepts, and CBT to promote deeper self-awareness and emotional clarity. The sequence in Fig- ure 8takes the client on a user-centred journey through chatting and journaling. This foundation builds user confidence and trust in effective therapeutic channels through visual synthesis. Clients interactively disambiguate cultural contexts and better articulate their values and struggles. This helps therapy reduce stigma and mitigate cultural disconnect. 6.3.1AI Chatbot The AI chatbot component of YAQIN facilitates personalised, supportive, and cul- turally sensitive dialogue. The prototype initially usesLangChainandOpenAIâs GPT-4.1-nano[ 65,66]. The early version, uses aPydanticschema to enforce struc- tured outputs [67]. The response consists of four segments: (i) empathetic reflection, (i) psycho-educational insight, (i) relevant reference from accessible and compre- hensive Quran exegesis rich with narration and philosophy [68], and (iv) a reflective question to promote introspection. This ensures cultural appropriateness while adhering to evidence-based therapeu- tic models. For users with negative past therapy experiences, this offers psycholog- ical safety and spiritual validation. This project iteratively tested and refined the 20YAQIN (a)(b)(c)(d) Figure 8: Concept UI Screens for (a) YAQIN Chatbot Landing Screen, (b) Example Entry and Response Chat, (c) Main Tafakkur (Reflection) Journal Landing Screen, and (d) Therapist Per- mission Request Feature chatbot tone and flow to treat faith-based constructs as sources of resilience and to eliminate any pathologising hallucinations. The project made the initial AI chat- bot, without RAG, on for evaluation. Figures 8(a, b) show the main interface and structured response concept, respectively. Design specifications met:1, 2, 3, 6, 9, and 10. 6.3.2Retrieval-Augmented Generation Pipeline RAG is a neural architecture that restricts the generative abilities of large language models (LLM) to answer from knowledge retrieved from domain-specific curated datasets. It presents effective and eïŹicient inference time-enhancement to the factu- ality and contextual alignment of LLM responses [ 69]. To enhance the relevance and coherence of chatbot responses over time, the pro- duction version of YAQIN incorporates a RAG pipeline. This allows theagentic chatbotto learn from user journaling data, thereby producing responses that reflect emotional continuity and personal relevance. The pipeline operates through four modules: (i) semantic embedding of journal en- tries (usingSentenceTransformers) [70], (i) indexing and retrieval through aFAISS vector store [ 71], (i) relating to chat history and context throughLangChain[65] , and (iv) response generation using aGPT-based LLM, [66]. This allows the re- sponse to address the present concern in light of the previous concerns and growth factors. For example, it responds to user anxiety through including successful inter- actions in past journal entries. Figure9illustrates this concept through an app-flow breakdown of a use case for the chatbot feature. This contextual agentic AI navigates client long-term journeys and simulates re- sponsive and culturally competent therapists, allowing continuous theological and Yasmin Zaraket21 Figure 9: YAQIN Chatbot Use Case Flow Figure 10: YAQIN RAG Diagram from the Use Case Flow spiritual framing. Design specifications met:1, 2, 5, 6, 9, 10, and 11. 6.3.3Tafakkur (Reflection) Journaling TheTafakkur(Reflection) Journal is the emotional heart and cognitive scaffold of YAQIN. It provides a space to record thoughts via text or voice, and automatically 22YAQIN curates entries by themes such as âfaithâ and âfamilyâ for easier retrospective access. The nameTafakkur(ààŹČÙÚ°) is Arabic for deep, contemplative reflection. In Islamic epistemology,tafakkuris a Quranic spiritual act to attain insight, self-awareness, and connection with Godâs manifestations. Selecting Tafakkur to name the journal feature grounds it in a culturally and theologically resonant practice, reinforcing YAQINâs goal to centre faith-sensitive introspection as a legitimate and valued path to emotional clarity and healing. Compared to other journaling apps, YAQIN uniquely foregrounds Islamic modes of reflection and addresses a clear market gap in culturally and spiritually tailored mental health tools [ 72]. YAQIN designs assistive prompts based on cognitive be- havioural principles and spiritual ethics. They help clients explore thoughts, reduce pressure to entry, and guide clarity and disambiguation without compromising val- ues. Figure8(c) shows the concept Tafakkur interface with selectable categorised and pre-prompted journal entries, and Figure11illustrates the app-flow breakdown. Design specifications met:2, 4, 5, 11, and 12. Therapist Integration Features In response to the limitations of existing therapy models for culturally diverse clients and the importance of accessing and attending therapy sessions with human therapists, YAQIN includes optional therapist integration features. YAQIN does not replace therapists, it supports them by enhancing culturally-sensitive mental healthcare services. Clients can share selected journal entries or auto-generated summaries with their therapists. This supports continuity, bridges cultural gaps, and reduces emotional burden on clients. Figure 11shows therapist notifications resulting from this feature illustrated in Figure8(d). YAQIN also facilitates approved therapists access to thematically sorted client reflections. Figure8(d) shows user permission request to share a client journal entry with their therapist. This adds client agency and optional privacy layer. This feature enables therapists to better prepare for sessions and to engage with client worldviews on theirs own terms regardless of cultural background. Design specifications met:1, 7, and 12. 7Evaluation This section presents Descriptive Study I, which informs refinements to YAQIN, shaping design specifications, user flows, and prototypes based on preliminary user research and cultural theory. The co-design session then validates and refines them, fostering collaborative ideation where participants shape how YAQIN might evolve to better reflect their needs. 7.1Objective This project structures the evaluation around two research questions that aim to further develop YAQIN: RQ1: How might Muslim women perceive and respond to a culturally sensitive mental health chatbot? Yasmin Zaraket23 Figure 11: YAQIN Journal Use Case Flow This evaluates whether the chatbotâs tone, content, and cultural framing emo- tionally resonates with the intended users. Given historical distrust and cultural mismatch in conventional mental healthcare, this project explores user perception evaluation as a critical first step to validate the interventionâs relevance and usabil- ity. RQ2: What co-designed insights can inform YAQINâs refinement to better support Clientsâ emotional and spiritual needs? This question aims to uncover how participatory input can refine the final iteration to meet real user needs. Co-designed feedback enables refinements in tone, reflective prompting, and spiritual guidance features that are key to building communication confidence and therapeutic readiness. 24YAQIN Both RQ1 and RQ2 support the main research question in Section3: to ex- plore whether a culturally and spiritually aware AI tool can facilitate emotional self-expression and act as a meaningful bridge to therapy for Muslim women in the UK. 7.2Methodology Health design research increasingly adopts co-design to include users in shaping in- terventions. It reveals tacit knowledge and culturally embedded expectations often missed in standard usability testing. As noted in recent health service design liter- ature, co-design enables the creation of more relevant, acceptable, and sustainable interventions by aligning outcomes with lived realities and values of users [73]. This was particularly relevant for YAQIN, where cultural sensitivity and theolog- ical alignment were essential design parameters. YAQIN validated and refined the âlooks-likeâ and âworks-likeâ prototypes based on co-design session findings, result- ing in direct user-led feedback on both content and delivery of the AI chatbot, which could be integrated into prompt engineering, RAG content selection, and prototype structure. The co-design session included four Muslim women with prior experience in mental health services. This number aligns with Nielsen Norman Group research showing that five users can uncover most usability issues [ 74]. In the context of co-design, smaller groups are methodologically appropriate and preferred because they fos- ter psychological safety and enable deeper, more emotionally nuanced engagement. YAQIN reinterviews an expert separately as the fifth participant to keep the co- design session client-focused while integrating professional feedback. 7.2.1Co-design session This project conducts the co-design session with four Muslim women in person, structured to include both individual and collaborative interaction with the chatbot. The session lasted approximately45minutes. Figure12outlines the session plan. The project exposed the clients to a working prototype of the chatbot deployed on Hugging Face Spaces instead of the development VSCode version. This focused the feedback on content and user experience while preserving performance speed, structured output parsing, and mirrored real-world deployment. The prototype offered intuitive, browser-based interaction, mirroring real-world use [ 75]. 7.2.2Feedback Synthesis Notes taken throughout the session captured verbal reactions and suggestions. This project summarises these notes to focus on analysing recurring patterns of emotional resonance, perceived relevance of spiritual content, trust in tone, and refinement suggestions. Key feedback on the chatbotâs design prompt directly shaped the updated sys- tem. P3 appreciated how the chatbot explained Quranic verses in a spiritually and emotionally meaningful way: Yasmin Zaraket25 Figure 12: Co-Design Session Plan âI loved that it comes with an explanation rather than just throwing a verse or verse number at me [...] it was [...] meaningful.â (P3) Participant 1 (P1) requested more practical suggestions beyond emotional com- fort: âI can imagine more lifestyle and routine support such as built-in check- lists and ideas on how to organise your day.â (P1) Participant 2 (P2) praised the reflection prompts but asked for more intention behind them: âThe question could be a bit more growth-focused [...] maybe guide me toward what I want to improve, not just reflect.â (P2) Participant 4 (P4) warned of accessibility and tone issues: âIt doesnât feel like a robot talking to you. Itâs like an older sibling giving advice [...] but to work for everyone, maybe it could read responses aloud or support more languages.â An expert interviewed in Descriptive Study I provided additional feedback. After testing the chatbot, she noted: 26YAQIN Figure 13: Prompt Refinement Overview Based on Co-Design â[I appreciated] the combination of summarising what I wrote through validation and the explanations to each response and linking to a verse from Quranâ (E5) She also emphasised how clients could benefit from using chatbot to prepare for therapy sessions. â[Thereâs benefit in] the fact that when you open up, you get a response that helps you understand yourself betterâ (E5) Her insights further validated the chatbotâs potential to support emotional pro- cessing and self-awareness ahead of clinical engagement and consequently improving the session. 7.3Findings The co-design session produced several key findings: âąEmotional and Spiritual Resonance:Participants described the chatbot as warm, spiritually aïŹirming, and emotionally safe. âItâs a wholesome response, very caringâ (P4) âItâs meaningful [...] like I may know the verse exists, but I donât always remember it, and this brings it back with explanation.â (P3) âąTherapeutic Utility:All participants agreed that the chatbot could be used to prepare for therapy or complement reflection work outside of sessions. âWhen youâre typing things out, itâs almost a way of being able to logically go through what youâre feeling and identify exactly what you need to speak about.â (P2) âI could probably ask the chatbot to help me explain something when my own words fail me.â (P3) Yasmin Zaraket27 âąDesign Improvements:YAQIN uses these and other insights to build an agentic RAG based system that directs the output of the LLM. Figure13also details the prompt refinement changes in red and green based on end-user feedback. Improvements included: simpler, directly relevanttafsir(exegesis); action- oriented questions; and integrated practical suggestions. Participants also proposed features like audio support, multilingual options, and deeper verse analysis. 7.4Final Design Iteration The findings and validation of the core design assumptions in YAQIN provided grounded directions for refinement. Critically, emotional safety, cultural fit, and spiritual framing of YAQIN made it a care tool and a therapy bridge. It fulfills the role of a culturally sensitive intervention in a digitally-mediated therapeutic landscape. The final YAQIN design, summarised in Figure14, illustrates how main features align with core and secondary design specifications. The dashboard includes wid- gets for Quranic framing and recurring themes from usersâ interaction like family, identity, and mood trends, aligning with the need for gentle, faith-based reminders and emotional tracking [ 73]. The updated chat-preferences panel satisfy co-design suggestions for greater agency and cultural personalisation in tone, religious fram- ing, and multi-language response options. Participants expressed interest in setting their âfaith-integration level,â allowing responses to be light, moderate, or deeply infused with Islamic content, mirroring P3âs feedback for âan option to go deeper intotafsir.â The Tafakkur Journal incorporates self-narration and cultural anchoring, reflect- ing feedback from users who found journaling to be emotionally clarifying but wanted the option to frame their responses in a faith-aware way. General-purpose AI models are predominantly trained on Western datasets, limiting their cultural nuance [ 76,77], RAG enables YAQIN to respond from a curated, context-specific knowledge base. This helps mitigate cultural mismatch by grounding responses in values and language aligned with the lived experiences of its users by generating context-aware responses grounded in prior journal entries and the select database of Islamic texts, enhancing emotional continuity and theological relevance. 28YAQIN Figure 14: Final Designed Main Features and their Alignment to the Design Specifications Additionally, this project updates the chatbot hosted on Hugging Face to simulate a working RAG technology with a database of Islamic texts and example journal Yasmin Zaraket29 Figure 15: Final User Journey Maps for (top) Muslim Clients MP1 and MP2, and (bottom) Therapists TP1 and TP2, Showing the Improved Interaction Pathways with YAQIN entry from MP2 Lina. The system uses Pinecone [78] to retrieve relevant user-specific context from this database before sending it with the prompt and user query to the LLM. To protect user privacy, individuals access the chatbot either as a gust or under the pseudonym âLina,â allowing the system to only reference the appropriate journal entries. Figures 15(top) and (bottom) present the improved use-case for all personas through iterated user journey maps from Section5. For Muslim clients MP1 and MP2, YAQIN supported clearer emotional expression, spiritual aïŹirmation, and ther- apy preparation. For therapists TP1 and TP2, the app offered culturally grounded insights to their clients that helped bridge value gaps after implementation in the session. 8Discussion This discussion explores the key strengths and limitations of the YAQIN prototype, its broader implications in faith-centred design, and methodological reflections on conducting sensitive, identity-based research. 30YAQIN 8.1YAQIN Strengths & Weaknesses YAQIN stands out as culturally-sensitive and responsive as it integratestawakkul (leaving matters in Godâs hands),sabr(patience), andtafakkur(reflection) Islamic concepts. Clients found its structured responses emotionally validating and spiritu- ally aligned: They felt seen and supported, unlike some of their past experiences with cultural dissonance in conventional therapeutic contexts. YAQIN with RAG allows recall and contextualisation of journal entries, creating a sense of therapeutic continuity. Initially, YAQIN used FAISS for vector similarity search. YAQINâs final iteration transitioned to OpenAIâs embeddings to enhance semantic understanding and accuracy. The critical healthcare nature of YAQIN justifies higher cost and requires accurate cross-lingual capture of emotional and cultural nuances [79]. However, limitations emerged. LLMs carry risks of confirmation bias and hallu- cinations which might produce misinterpretations of distress and vague responses. While YAQIN was emotionally supportive, clients desired more structured advice grounded in Islamic ethics navigating complex life decisions. This project plans to involve experts and scholars to refine guidance theologically and address these gaps. 8.2Broader Impact & Contribution YAQIN contributes to society practically by ethically and contextually aligning AI with the needs of underserved religious communities. Its hybrid use of psychology- informed language and spiritual framing offers a blueprint for future tools that integrate therapists and complement rather than replace expert healthcare. It also contributes through faith-sensitive human-computer interactions by opera- tionalising Islamic frameworks within mental healthcare design. It extends cultural competence by integrating theological literacy and Islamic epistemologies into CBT. This forges novel culturally-sensitive practices that leverage technology to culturally inform and engage existing conventional practices. It urges designers to prioritise lived values, relational ethics, and social identities as inclusive innovation founda- tions. 8.3Research Methodology Evaluation & Limitations The DRM offered a structured yet flexible framework for this iterative project. Semi- structured interviews captured nuanced experiences. Deductive thematic analysis grounded insights in literature. The co-design session crucially shaped YAQINâs emotional tone and faith and cultural-sensitive approach. This project has few limitations. It involved a small, geographically limited sample of mostly UK-based, middle-income university students in their early twenties with regular digital access. This homogeneity restricts generalisability. Broader cultural and international representation could strengthen applicability. Although ethical safeguards were in place, exploring mental healthcare may pose emotional risks, requiring continued sensitivity in future research. For the scope of this project, the focused sample allowed for deep engagement Yasmin Zaraket31 with the target user group, generating valuable insights and design principles. This project used these findings to inform development while contributing meaningful knowledge on how faith-sensitive tools can be thoughtfully adapted and implemented in other cultural and religious contexts. 8.4Future Implications of Findings This study offers a scalable model for culturally-competent digital design. While tailored for Muslim women, YAQINâs principles could extend to other faith and cultural groups. The adaptable structure of YAQIN supports this cross-context application, provided it is redesigned with appropriate accounting for age, digital fluency, and theological and cultural variations. YAQIN mitigates generic or irrelevant reflections through its use of RAG, anchor- ing responses in the client journal entries and curated Islamic content. However, there remains a risk of misinterpretation or unmet expectations. Future iterations of YAQIN should refine disclaimers, increase feedback loops, and diversify testing to ensure psychological safety, generalisability, and scalability. Findings from this project reinforce the need for deeper therapist training in religious literacy and culturally responsive care in clinical and educational contexts. As AI becomes more prevalent in mental healthcare, it is vital that diverse world- views are adequately and appropriately included. YAQIN shows how innovative faith-sensitivity and cultural awareness effectively support healing and sets a novel precedence in inclusive digital mental healthcare. 9Conclusion YAQIN uses culturally tailored agentic AI to address persistent gaps in mental healthcare for Muslim women in the UK. It followed a DRM approach and was co- designed with stakeholders to integrate a culturally-sensitive AI chatbot, journaling tools, and therapist access, all grounded in contemporary culturally inclusive psycho- logical frameworks. This bridges a mismatch between under-resourced conventional approaches and Muslim women therapy expectations. The findings from user research, thematic analysis, and co-design sessions vali- date the potential of YAQIN to foster emotional trust, enhance communication con- fidence, and support pre-therapeutic reflection. Participants felt heard, respected, and spiritually aïŹirmed by the reflections. This confirms YAQINâs promise as a bridge to culturally and faith-sensitive mental healthcare. Nonetheless, limitations exist. The sample size was relatively small and limited to UK-based participants, restricting the generalisability of findings to other pop- ulations or cultural contexts. The tone and theological framing of YAQIN were positively received. Yet, future work is needed to evaluate long-term engagement and mental healthcare outcomes in clinical practice. Future directions for YAQIN include: âąConduct longitudinal study with diverse communities to test generalisability and sustained impact 32YAQIN âąContinuously refine and reinforce the AI model with relevant theological and psychological knowledge and user feedback âąExplore piloting opportunities in collaboration with mental healthcare providers and community centres This project lays foundations for a new class of culturally-sensitive agentic AI. With further refinement, YAQIN can become a scalable intervention that supports Muslim women and redefined inclusivity in healthcare. As generative AI continues to develop, this shows how AI can embed ethical, spiritual, and cultural values in human-centred design. Yasmin Zaraket33 References [1]T. Bignall, S. Jeraj, E. Helsby, and J. Butt, âRacial disparities in mental health: Literature and evidence review,â tech. rep., Race Equality Foundation, London, UK, 2020. [2]OïŹice for National Statistics, âReligion by age and sex, england and wales: Census 2021,â 2023. [3]D. Kapadia, J. Zhang, S. Salway, J. Nazroo, A. Booth, N. Villarroel-Williams, L. BĂ©- cares, and A. Esmail, âEthnic inequalities in healthcare: A rapid evidence review,â tech. rep., NHS Race and Health Observatory, 2022. [4]A. Haque, F. Khan, H. Keshavarzi, and A. E. Rothman, âIntegrating islamic tradi- tions in modern psychology: Research trends in last ten years,âJournal of Muslim Mental Health, vol. 10, no. 1, 2016. [5]British Medical Association, âMental health pressures data analysis,â 2025. [6]Muslim Youth Helpline, âMuslim youth helpline,â 2025. [7]Inspirited Minds, âInspirited minds,â 2025. [8]Muslim Womenâs Network UK, âMuslim womenâs network uk,â 2025. [9]N. Bansal, S. Karlsen, S. P. Sashidharan, R. Cohen, C. A. Chew-Graham, and A. Malpass, âUnderstanding ethnic inequalities in mental healthcare in the uk: A meta-ethnography,âPLOS Medicine, vol. 19, no. 12, 2022. [10]Muslim Mind Collaborative, âMental health in uk muslim communities: A call to action,â tech. rep., Muslim Mind Collaborative, 2024. [11]F. Ahmad, A. Khan, I. Patel, and Y. Rahman, âReview of muslim patient needs and its implications on healthcare delivery,âJournal of Health Care for the Poor and Underserved, vol. 34, no. 1, 2023. [12]City St Georgeâs, University of London, âMuslim mental health matters report,â 2022. [13]K. Bhui, âCulture and mental health: a comprehensive textbook,âInternational Re- view of Psychiatry , 2007. [14]E. Eustis, H. Rhoades, N. Kanuri, V. Narayanan, C. Gregor, and D. Kaysen, âCultural responsivity in technology-enabled mental health services: A conceptual framework,â Journal of Technology in Behavioral Science, vol. 8, no. 3, p. 475â491, 2023. [15]S. Thakkar, M. Kharma, and C. El Morr, âArtificial intelligence for positive mental health: A narrative review,âArtificial Intelligence in Medicine, vol. 151, p. 102885, 2024. [16]M. Hossain, M. Alam, T. Rahman, and M. Abedin, âSmile: A culturally sensitive ai-powered mental health platform for global deployment,âComputers in Biology and Medicine, vol. 184, p. 109240, 2025. [17]L. Zhang, P. Mwangi, and S. HernĂĄndez, âCross-cultural validation of ai-assisted mental health screening tools: Challenges and opportunities,âFrontiers in Digital Health, vol. 3, p. 144â162, 2025. 34YAQIN [18]K. M. Almutairi, V. L. McCarthy, and G. Gardner, âIraqi muslim women and primary healthcare providersâ perspectives and experiences in the clinic visit in the western united states,âQualitative Health Research, vol. 32, no. 2, 2022. [19]J. Sherif, âThe impact of unconscious bias on muslim womenâs experiences of health- care,âBJGP Life, 2021. [20]S. Elahi, D. Elsayed, S. Ali, and R. Awaad, âEthnic differences in muslim womenâs mental health beliefs, rejection attitudes, and familiarity with professional mental healthcare,âMental Health, Religion & Culture, vol. 26, no. 9, 2023. [21]A. Tannerah, O. Hazel, S. Desson, R. Farah, Z. Kamil-Thomas, H. Iqbal, C. Eames, P. Saini, and O. Bifarin, âConsultations with muslims from minoritised ethnic com- munities living in deprived areas: Identifying inequities in mental health care and support,âHealth Expectations, vol. 27, no. 4, 2024. [22]F. Z. Hussain, âMental health attitudes and knowledge among shia muslims in amer- ica,â honors undergraduate thesis, University of Central Florida, Orlando, FL, 2022. [23]A. Mahmud, âAddressing inequalities in accessing mental health services for muslim university students in the united kingdom: Implications for attainment and wellbe- ing,âJournal of Beliefs & Values, vol. 45, no. 1, 2024. [24]Y. Chen and W. Mak, âUnderstanding and addressing mental health stigma across cultures,âInternational Journal of Mental Health, vol. 31, no. 2, 2023. [25]S. Ugar and I. Malele, âDesigning ai for mental health diagnosis in sub-saharan africa: Why cultural context matters,âJournal of Medical Ethics, vol. 50, no. 9, p. 592â598, 2024. [26]H. Lee, A. Alabdulkarim, and C. Webster, âCultural prompting improves empathy and appropriateness in ai mental health agents,âACM Transactions on Computer- Human Interaction, vol. 32, no. 1, p. 1â29, 2025. [27]S. Ali, J. Smith, and A. Khan, âTowards a better understanding between non-muslim primary care providers and muslim patients,âJournal of Health Communication, vol. 12, no. 1, 2023. [28]L. Rahman, R. Patel, and S. Ahmed, âMental health care for muslims: Can it be person-centered without considering religious identity?,âTranscultural Psychiatry, vol. 59, no. 4, 2022. [29]Z. Saherwala, S. Bashir, and D. Gainer, âProviding culturally competent mental health care for muslim women,âInnovations in Clinical Neuroscience, vol. 18, no. 4- 6, 2021. [30]G. M. Spreitzer, âPsychological empowerment in the workplace: Dimensions, mea- surement, and validation,âAcademy of Management Journal, vol. 38, no. 5, 1995. [31]A. Radovanovic, L. Daly, J. Mitchell, and R. Kline, âCo-designing a chatbot for racial and cultural competence in clinical communication training,âPatient Education and Counseling, vol. 132, p. 107â118, 2025. [32]S. C. R. Team, âUsing ai-enabled deliberate practice to improve cultural competence in behavioural health clinicians,âBehavior Therapy, vol. 55, no. 4, p. 555â568, 2024. Yasmin Zaraket35 [33]J. Rand, R. Khosla, and A. Malik, âCulturally appropriate care in the age of ai: Eth- ical challenges and opportunities,âJournal of Medical Ethics, vol. 51, no. 2, p. 112â 119, 2024. [34]S. Abrar and J. Hargreaves, âMental health services for muslim com- munities in england and wales: developing a more collaborative model.â https://w.cis.cam.ac.uk/wp-content/uploads/2024/02/Abrar-and-Hargreaves- 2023-Mental-health-services-for-Muslim-communities.pdf, 2023. [35]H. Raza and S. Ahmed, âMindfulness in salah prayer and its association with mental health,âJournal of Religion and Health, vol. 58, no. 3, 2019. [36]J. Owens, G. H. Rassool, J. Bernstein, S. Latif, and B. H. Aboul-Enein, âInterventions using the qurâan to promote mental health: A systematic scoping review,âJournal of Mental Health, vol. 32, no. 4, 2023. [37]S. F. Ahmad, H. Han, M. M. Alam, M. K. Rehmat, M. Irshad, M. Arraño-Muñoz, and A. Ariza-Montes, âImpact of artificial intelligence on human loss in decision making, laziness and safety in education,âHumanities and Social Sciences Communications, vol. 10, no. 1, 2023. [38]R. Tampubolon and B. Nadeak, âArtificial intelligence and understanding of religion: A moral perspective,âJournal of Social and Religious Research, 2024. [39]J. M. Smyth, K. E. Heron, S. A. Wonderlich, M. J. Sliwinski, R. D. Crosby, and J. E. Mitchell, âOnline positive affect journaling in the improvement of mental distress and well-being in general medical patients with elevated anxiety symptoms: A preliminary randomized controlled trial,âJMIR Mental Health, vol. 5, no. 4, p. e11290, 2018. [40]H. Gardiner and N. Mutebi, âAi and mental healthcare â opportunities and delivery considerations,â Tech. Rep. POST-PN-0737, UK Parliamentary OïŹice of Science and Technology, 2025. [41]T. Nadarzynski, N. Knights, D. Husbands, C. Graham, C. D. Llewellyn, T. Buchanan, I. Montgomery, A. S. Rodriguez, and C. Ogueri, âChatbot-assisted self-assessment (casa): Co-designing an ai-powered behaviour change intervention for ethnic minori- ties,âPLOS Digital Health, vol. 4, no. 2, 2025. [42]S. Tunggala, âCross-cultural communication in the era of ai: An islamic ethical frame- work,âSocieto Communication Journal, vol. 2, January 2025. [43]Calm Health, âCalm health â mental health solutions for organizations,â 2025. [44]Headspace, âHeadspace â meditation and sleep made simple,â 2025. [45]Wysa, âWysa â ai mental health support,â 2025. [46]Muslim Moodfit, âMuslim moodfit â mental health app,â 2025. [47]Ruhi, âRuhi â islamic meditation app,â 2025. [48]Inspirited Minds, âInspirited minds â faith-based mental health charity,â 2025. 36YAQIN [49]F. Ibrahim, A. Alsudais, Y. Yao, and I. Shklovski, âTracking during ramadan: Ex- amining the intersection of menstrual and religious tracking practices among muslim women in the united states,â inCHI Conference on Human Factors in Computing Systems, ACM, 2024. [50]R. Hadfi, Y. Kitamura, and T. Okamoto, âConversational agents enhance womenâs contribution in online debates,âScientific Reports, vol. 13, no. 1, p. 1â14, 2023. [51]E. Emke, âThe risk of cultural insensitivity in ai translations,â December 2024. [52]R. Dehbozorgi, S. Zangeneh, E. Khooshab, D. Hafezi Nia, H. R. Hanif, P. Samian, M. Yousefi, F. Haj Hashemi, M. Vakili, N. Jamalimoghadam, and F. Lohrasebi, âThe application of artificial intelligence in the field of mental health: A systematic review,â BMC Psychiatry, vol. 25, no. 132, 2025. [53]E. Johnson and M. Lee, âArtificial intelligence in healthcare: A critical analysis of the ethical implications,âJournal of Medical Ethics, vol. 50, no. 3, 2024. [54]University of Westminster, âUniversity of westminster study pioneers effective chat- bot self-assessment questions to tackle health inequalities among ethnic minorities,â February 2025. [55]J. Nelson, A. AbuHaimed, and R. Zhang, âDesigning ai systems for diverse cultural contexts,âAI and Society, 2025. [56]L. T. Blessing and A. Chakrabarti,DRM: A Design Reseach Methodology. London: Springer London, 2009. [57]S. Kvale and S. Brinkmann,InterViews: Learning the craft of qualitative research interviewing. SAGE Publications, 3rd ed., 2015. [58]G. Guest, E. Namey, and L. Chen, âHow many interviews are enough? a test of data saturation and variability,âField Methods, vol. 32, no. 1, 2020. [59]D. Byrne, âA worked example of braun and clarkeâs approach to reflexive thematic analysis,âQuality & Quantity, vol. 56, no. 3, 2022. [60]A. Blomquist and M. Arvola, âPersonas in action: ethnography in an interaction design team,â 2002. [61]Yale University Usability Group, âUser journey maps,â 2023. [62]R. Hartson and P. Pyla, âUser-centered design: Personas and scenarios,â inThe UX Book, Springer, 2019. [63]OpenAI, âChatgpt: Language model (gpt-4).âhttps://chat.openai.com, 2023. ChatGPT used to generate persona images and proofread report for spelling and grammatical errors. [64]Design & Paper, âMuted color palettes for a peaceful state of mind,â 2021. [65]H. Chase, âLangchain: Building applications with llms through composability,â 2022. [66]OpenAI, âGpt-4.1-nano language model.âhttps://openai.com, 2024. Yasmin Zaraket37 [67]S. Colvin, âPydantic: Data validation and settings management using python type annotations,â 2021. [68]A. Tabatabaâi, âTafsir al-mizan â an exegesis of the holy quran by allamah tabatabai.â [69]P. Lewis, E. Perez, A. Piktus, F. Petroni, V. Karpukhin, N. Goyal, I. Kulikov, and et al., âRetrieval-augmented generation for knowledge-intensive nlp tasks,âAdvances in Neural Information Processing Systems, vol. 33, p. 9459â9474, 2020. [70]N. Reimers and I. Gurevych, âSentence-bert: Sentence embeddings using siamese bert-networks,âarXiv preprint arXiv:1908.10084, 2019. [71]J. Johnson, M. Douze, and H. JĂ©gou, âBillion-scale similarity search with gpus,â 2017. [72]A. Rashid, âThe rise of faith-centric mental health apps in muslim communities,â 2023. [73]H. Christie, S. Mahmood, and R. Malik, âCodesigned, faith-informed interventions in mental health: Lessons from muslim and bame communities,âJournal of Partici- patory Health Research, vol. 4, no. 1, 2022. [74]N. N. Group, âWhy you only need to test with 5 users,â 2000. [75]H. Face, âHugging face spaces.âhttps://huggingface.co/spaces, 2021. [76]E. M. Bender, T. Gebru, A. McMillan-Major, and S. Shmitchell, âOn the dangers of stochastic parrots: Can language models be too big?,â inProceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, 2021. [77]A. Abid, M. Farooqi, and J. Zou, âPersistent anti-muslim bias in large language models,â inProceedings of the 2021 AAAI/ACM Conference on AI, Ethics, and Society, 2021. [78]Pinecone Systems Inc., âBuild a rag chatbot,â 2025. [79]OpenAI, âNew embedding models and api updates,â January 2024.