Speech And Language Signals
Using spoken and written language to study anxiety, depression, and symptom-relevant clinical interviews.
Applications of AI in mental health care
Bazen Gashaw Teferra, PhD
I study how speech, language, and large language models can support mental health screening, symptom monitoring, and care navigation while staying safe, interpretable, equitable, and clinically accountable.
Research identity
Building mental health AI that can be evaluated before it is trusted.
Dr. Bazen Gashaw Teferra is an AI and mental health researcher in Toronto working across computer engineering, natural language processing, psychiatry, digital health, and equitable health service delivery. He is part of the AI for Mental Health research group. His work asks how language technologies can help identify and monitor mental health needs without replacing clinicians or weakening clinical accountability.
His research has moved from acoustic and linguistic signals of generalized anxiety disorder to transformer-based screening, LLM-based depression assessment, therapeutic communication evaluation, safety guardrails, agentic AI in psychiatry, and fairness auditing.
LLM Safety And Evaluation
Testing hallucination, false reassurance, refusal behavior, empathy, crisis response, and false-premise agreement.
Equity And Governance
Evaluating subgroup performance, demographic bias, consent, privacy, and community-defined harm reduction.
Clinical Translation
Designing clinician-facing workflows, benchmarks, summaries, and implementation guardrails for real care settings.
Selected work
Projects at the intersection of psychiatric measurement and AI safety.
Published research
LLM-Based Depression Screening
Evaluating whether LLMs can infer item-level PHQ-8 symptoms from clinical interview transcripts.
Read the paperPublished research
Safety Guardrails And Irritability
Studying how safety guardrails shape affective realism and therapeutic communication risk in LLMs.
Read the paperActive manuscript
PHQ-8 Demographic Bias Audit
Auditing LLM depression-screening outputs across gender and age subgroups using patient-level error metrics.
Active experiment
False-Premise Resistance
Testing whether single-agent and multi-agent LLM systems can resist pressure to agree with inaccurate or unsafe premises.
Public code
Therapeutic Alliance Simulation
Simulating LLM patient-therapist conversations and evaluating therapeutic alliance style signals.
View repositoryPublic code and tools
Selected GitHub repositories
LLM_v_LLM_WAI
Therapeutic alliance simulation and WAI-style evaluation.
TherapySimulation
LLM therapist-patient conversation simulation and evaluation tooling.
LLM_Irritability
Code connected to safety guardrails and irritability metrics.
medicalSummarizer
FastAPI service for summarizing medical text with LLMs.
PTSD_screening
Notebook-based PTSD screening and PCL-C processing work.
More on GitHub
View the full public repository profile.
Independent technical work
Personal engineering projects outside my formal research portfolio.
Free-time project
Amharic Speech-To-Text
A personal language-technology project focused on Amharic automatic speech recognition, including Whisper fine-tuning, Common Voice and dialect-aware data preparation, transcription API design, and practical evaluation with WER/CER metrics.
Publications
Peer-reviewed publications and public scholarly outputs.
Alignment of Large Language Model Responses With Human Therapists in Motivational Interviewing
JAMA Network Open, 2026
Assessing the impact of safety guardrails on large language models using irritability metrics
npj Digital Medicine, 2026
Reimagining psychiatric care with agentic AI: promise, challenges, and a roadmap forward
npj Digital Medicine, 2026
Leveraging large language models for automated depression screening
PLOS Digital Health, 2025
Opportunities and Barriers of Generative Artificial Intelligence in the Training of Psychiatrists
Academic Psychiatry, 2025
Screening for Depression Using Natural Language Processing: Literature Review
Interactive Journal of Medical Research, 2024
Predicting Generalized Anxiety Disorder From Impromptu Speech Transcripts
JMIR Mental Health, 2023
Screening for Generalized Anxiety Disorder From Acoustic and Linguistic Features
JMIR Formative Research, 2022
Acoustic and Linguistic Features of Impromptu Speech and Their Association With Anxiety
JMIR Mental Health, 2022
Experience
Engineering training, clinical AI research, and community-facing service.
Postdoctoral Fellow, St. Michael's Hospital, Unity Health Toronto
AI for mental health, LLM-based assessment, digital psychiatry, safety guardrails, and clinician-in-the-loop workflows.
PhD, Computer Engineering, University of Toronto
Doctoral research on generalized anxiety disorder prediction from acoustic and linguistic features of impromptu speech.
Teaching And Research, Addis Ababa University
Computer engineering teaching, supervision, and applied student project mentorship.
Community And Health Equity Service
Board service with People To People Canada and prior board service with the Ethiopian Diabetes Association.
Community service
Community-facing work connected to health, education, and equity.
Board service
People To People Canada
Board member with a Canadian charity whose work includes support for orphaned and vulnerable children in Ethiopia, community health programs, preventive health engagement, and mental-health awareness and education initiatives.
Prior board service
Ethiopian Diabetes Association
Supported health-equity work for people living with diabetes in Ethiopia, including awareness, fundraising, and advocacy related to access to insulin for children with Type 1 diabetes.
Connect