Skip to content

[ask-a-sailor] Theory of Mind evaluation module — score empathy progress per session #18

Description

@fullharbor

Nature 2025 (PMC12501279): GPT-4o can evaluate autistic participants' responses to Theory of Mind tasks, matching human clinical psychologists. This means the AI can function as an evaluator, not just a coach. Work: (1) Create src/evaluation/tom_evaluator.py -- takes a conversation transcript, scores across ToM dimensions: perspective-taking, emotional recognition, social inference, (2) output a structured EvaluationResult: score per dimension (0-5), overall score, improvement delta from previous session, (3) store results in Supabase workspace_ai_chats table (add evaluation_score JSONB column), (4) expose via API GET /sessions/{id}/evaluation. This turns the product from a chat widget into a measurable intervention tool -- the key differentiator for practitioners and funders. Ref: PMC12501279, PHDT paper (89.8% social cue recognition baseline)

Metadata

Metadata

Labels

enhancementNew feature or request

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions