PhD in mental-health AI | Fair assessment of psychotic symptoms and co-designed use

PhD in mental-health AI | Fair assessment of psychotic symptoms and co-designed use

Quick Decision

Promising, but clarify funding and intended use. This Cardiff project focuses on whether large language models can assess psychotic symptoms in mental-health records fairly and reliably, where they fail, and how people with lived experience should shape responsible use. It suits applicants with a 2:1 degree or Master's in computing or another quantitatively strong field, and it combines NLP evaluation with clinical and participatory research. The project description gives an application deadline of 5 January 2027 and October 2027 entry. It identifies an EPSRC DLA competition but does not state the award value, stipend, fee coverage, or project-specific funding duration. Before investing heavily in an application, clarify whether the project has a funded award allocated, how international fee status is handled, what “deployment” means, and whether clinical data and lived-experience partners are confirmed.

Research Problem and Doctoral Work

Clinical records contain rich descriptions of symptoms and care, but much of this material is unstructured text. The project asks whether LLM-based assessment can make symptom information more usable for research or service improvement without hiding uncertainty or worsening disparities. Its stated work includes testing reliability, fairness and robustness; examining when errors occur; checking whether performance differs by clinical context or demographic group; and co-designing risk-reduction guidance with lived-experience perspectives. Human judgement and oversight remain part of the project’s stated scope.

This is a high-stakes evaluation and responsible-use project, not a confirmed autonomous diagnosis or treatment system. The opportunity does not name the models, clinical records, partner services, evaluation criteria, intended users, or the boundary between research-only use and clinical support. “Deployment” could mean a research workflow, service evaluation or a clinical-facing tool; those imply different evidence, safety and governance needs. The first conversation should establish that boundary and the decision the work is meant to inform.

There is a useful local research context. Cardiff announced the separate SYLFAEN-PSY programme, a Wellcome-funded collaboration involving researchers, lived-experience experts, NHS Wales, industry and mental-health charities. A 2026 Cardiff-affiliated preprint on symptom extraction from clinical notes also illustrates why domain-specific assessment matters: it reports limitations and unevenness across symptom tasks and cautions against assuming general-purpose models are ready for clinical use. Neither source establishes that this EPSRC PhD is part of that grant or will access its people, data or infrastructure. The preprint is not peer reviewed at posting.

School, Programme and Field Context

The host is Cardiff University’s School of Computer Science and Informatics; the advertised route is a PhD in Computer Science and Informatics, with October 2027 entry. The school page describes a normal full-time PhD duration of three years and progress monitoring every six months. Cardiff’s EPSRC Algorithmic Grant DLA describes 3.5 years of studentship support and identifies 2027/28 as its final cohort. These periods may refer to different things, but the project-specific registration and funding end dates are not stated and should be reconciled in writing.

Cardiff is ranked =179 overall in QS World University Rankings 2027; Times Higher Education places its Computer Science subject in the 201–250 band for 2026. These are broad university and subject signals, not measures of this project’s supervision or clinical research access. Cardiff rankings · THE Cardiff profile. Cardiff’s research environment crosses computer science and psychosis research, with the separate SYLFAEN-PSY collaboration offering a potentially relevant local context. Access to that collaboration should not be assumed. The school page describes research and professional development, but the project does not state a taught-course load or a dedicated clinical training component.

Supervisor and Lab Assessment

Dr Yi Zhou is a Lecturer in Computer Science and Informatics whose stated research includes NLP fairness and bias, multilingual language approaches, interpretability and model analysis. Dr Víctor Gutiérrez-Basulto is a senior computer-science academic working on knowledge representation, automated reasoning and knowledge-aware NLP. Together, those profiles suggest a plausible combination of fairness-oriented NLP evaluation and structured reasoning, but the project does not specify each supervisor’s day-to-day role or who will lead clinical and co-design work.

Three recent papers by Yi Zhou help locate the technical fit:

• Evaluating the Effect of Retrieval Augmentation on Social Biases (EACL, 2026): examines how retrieval augmentation can affect social bias in generated responses. It is directly relevant to fairness evaluation, but does not involve psychosis or clinical records. Paper.
• Evaluating Short-Term Temporal Fluctuations of Social Biases in Social Media Data and Masked Language Models (EMNLP, 2024): studies whether measured bias changes over time. Its findings underline that evaluation results depend on context and measurement choices; the models and data differ from the proposed clinical setting. Paper.
• A Predictive Factor Analysis of Social Biases and Task-Performance in Pretrained Masked Language Models (EMNLP, 2023): examines relationships between bias and task performance across pretrained models. It provides background for considering fairness alongside usefulness, but it is not evidence of clinical validity. Paper.

Recent work elsewhere shows that LLM symptom and risk assessment in psychosis is an active research area. A 2026 npj Digital Medicine study and a separate symptom-severity proof of concept provide adjacent evidence, but use different populations and tasks. These early studies do not settle generalization, safety or clinical utility. Cardiff’s own preprint and the supervisors’ fairness work make this PhD timely; its contribution will depend on whether it can obtain appropriate clinical and lived-experience input and define success beyond an average performance score.

Funding, Application, and Living Cost

The application instructions say that applicants are considered for a 2026 EPSRC DLA funded studentship when applying for the PhD, and direct them to select “No” for self-funding and enter: “I am applying for 2026 EPSRC DLA funded PhD Studentship in Computer Science and Informatics”. Cardiff’s scheme page describes 3.5-year support and nine awards each year across the university, with a final cohort starting in 2027/28. That does not confirm an award allocation for this project or guarantee funding to each applicant. The project description does not publish the stipend, exact length, fee coverage or research-cost allowance. The UKRI 2026/27 minimum stipend reference is £21,805 from October 2026; treat it only as a benchmark, not as a confirmed 2027/28 offer. Cardiff says successful UKRI-funded EPSRC awardees are not charged an international fee difference under its fee-discount policy, subject to the UKRI cap on international awards. Confirm that this specific competition has an international place and that the fee arrangement applies to it. DLA scheme · EPSRC studentship policy · UKRI stipend rates.

The stated deadline is 5 January 2027; the advert does not give a time or time zone. Apply online to the Computer Science and Informatics PhD and make clear in the research proposal field that the application is for this project, naming the supervisors and including the project description. The personal statement must be no more than 500 words and address research interests, preparation for PhD study and the project’s importance. Other materials include a CV, certificates and transcripts (with English translations if needed), two references and proof of English where applicable. The stated academic threshold is a 2:1 undergraduate degree or Master's in computing or a subject with substantial quantitative content; suitable professional experience is also considered. If IELTS is required, the listed threshold is 6.5 overall and 6.0 in each component. The strongest-fit candidates may be interviewed in person or remotely.

Cardiff’s 2026/27 calculator estimates £13,510.92 per year for a shared private four-bedroom accommodation scenario, including its listed routine costs but excluding travel. This is a planning estimate, not a promise of personal costs. Because the project stipend is not confirmed, the available information is insufficient to assess its buying power. Build a 12-month estimate that includes transport, deposits, visa and relocation costs, and ask whether any research/training allowance is separate from the stipend. Cardiff living-cost calculator (2026/27).

Applicant Fit and Preparation

Required background: a 2:1 honours undergraduate degree or a Master's in computing, or another subject with a significant quantitative component. The description names mathematics, engineering, economics and psychology as examples. Appropriate professional experience may also be considered. English-language proof is required where applicable.

Useful preparation: evidence of quantitative reasoning, NLP/LLM evaluation, bias or robustness analysis, and ability to communicate results across technical and clinical audiences. These are project-fit suggestions rather than additional formal eligibility criteria. Experience listening to and working respectfully with people who have relevant lived experience may help, but the advert does not specify a prior clinical qualification. Applicants should avoid claiming that a model can diagnose psychosis; the project is about evaluating limits and responsible use.

The combination can transfer to responsible AI, NLP research, digital-health evaluation, clinical informatics and academic research. Cardiff’s programme page describes research and development capabilities sought by employers, but there is no project-specific placement or salary evidence. Treat those as plausible pathways, not guaranteed outcomes.

Next Actions and Questions to Verify

Before Applying

1. Confirm whether the project has a funded EPSRC DLA place allocated, whether the instruction’s “2026” award wording applies to the October 2027 intake, the 2027/28 stipend, funding duration, home/international fee treatment, and any residency or international-place cap.
2. Ask what “co-designed deployment” means: research-only use, service evaluation or a clinical-facing system, and what decisions the research is meant to support.
3. Confirm whether clinical data, clinical collaborators and lived-experience contributors are already in place, and whether this PhD is linked to the separate SYLFAEN-PSY programme.
4. Prepare the required application materials and ensure the personal statement is within 500 words. Check the exact closing time and referee submission process with the School.

At Interview

1. Ask how supervisors divide technical, clinical and participatory responsibilities, and what regular input and support the student will receive.
2. Ask how the project will judge reliability, fairness and robustness across contexts, and how error analysis and human oversight influence the final guidance.
3. Clarify data access, ethics and privacy governance, the role and compensation of lived-experience contributors, and whether clinical-safety or regulatory advice is available if the intended use extends beyond research.

Before Accepting an Offer

1. Get the award terms in writing, including stipend, fees, duration, travel/research allowance, visa/relocation support and the relationship between the three-year programme duration and 3.5-year funding.
2. Compare the confirmed stipend with a full-year Cardiff housing and transport plan; the published shared-housing estimate excludes travel and will not fit every student.

Evidence and Sources

Project, School and Funding

• Computer Science and Informatics PhD programme
• Cardiff EPSRC Algorithmic Grant DLA
• Cardiff EPSRC studentship policy
• UKRI support and stipend rates
• Cardiff living-cost calculator 2026/27
• Cardiff rankings
• THE Cardiff University profile

Supervisors and Cardiff Research Context

• Yi Zhou profile
• Víctor Gutiérrez-Basulto profile
• Cardiff announcement of SYLFAEN-PSY
• Wellcome SYLFAEN-PSY grant record
• Cardiff-affiliated psychosis symptom extraction preprint
• EACL 2026 paper on retrieval augmentation and social bias
• EMNLP 2024 paper on temporal variation in social bias
• EMNLP 2023 paper on bias and task performance

Clinical LLM Evidence and Governance

• npj Digital Medicine study on LLMs and psychosis-risk assessment
• Proof-of-concept study on LLM symptom-severity measurement
• WHO guidance on AI ethics and governance for health
• NICE evidence standards for digital health technologies
• MHRA guidance on software and AI as a medical device