research
Make the evaluation question explicit
AI and mental-health research with evidence synthesis, prototype exploration, and human oversight.
- Role
- Research Assistant in AI and Mental Health
- Status
- Research role completed · June–September 2026
Context and real problem
Mental-health AI research needs clear distinctions between a research idea, a prototype, and a system appropriate for care. Technical capability alone does not establish clinical usefulness, safety, or permission to handle sensitive information.
Andres’s role
In a Research Assistant role in AI and Mental Health, I contributed to evidence synthesis, research planning, prototype exploration, and scientific communication. The recorded period is June to September 2026.
Collaborators and setting
The work was within a University of Toronto / Unity Health Toronto research collaboration. Clinical and technical contributors bring different expertise. This is a research role; it does not imply a clinical appointment or independent responsibility for treatment.
Methods and tools
The work includes literature searching, evidence synthesis, methodological documentation, and evaluation planning for language-model and agent workflows. Python, version control, and language-model tooling support prototype exploration.
Responsibility and safety
Clinical safety, transparency, privacy, bias, and appropriate human oversight belong in the evaluation plan. A prototype must have a bounded purpose. Research on a conversational or agent workflow is not evidence that it can autonomously assess or treat someone.
Outputs and outcomes
My contributions included review material, evaluation planning, prototype exploration, and scientific communication. This is research-stage work; clinical effectiveness and readiness for use in care remain unestablished.
What I learned
Evaluation is part of defining a system, not a final score attached after building it. I want the clinical question, possible harms, and decisions requiring human responsibility to shape the research from the beginning.
Current status and links
The role ran from June to September 2026. The linked professional profile provides further career context. Unpublished research detail remains within the research setting.