Big Five Conversational Assessment

Problem

A conversation can gather richer context than a static questionnaire, but extracting trait signals requires explicit state and careful interpretation.

Architecture

A FastAPI application connects survey routes, an interview session, structured LLM output, an embedding-based comparator, and SQLite persistence. Conversation history and facet coverage are maintained outside the model chain.

Implementation

The public agent uses a prompt, Groq-backed model, and Pydantic output parser. It tracks facet coverage and limits repeated probing. The axis scorer subtracts low-pole from high-pole exemplar embeddings, projects responses onto that direction, and maps the result to a bounded scale.

Results

VERIFIED by source inspection: external session state, structured parsing, contrastive exemplar scoring, and transcript persistence exist. A separately audited private revision is not published here and is not represented as a second project.

Limitations

Exemplar calibration is not population validation. No clinical diagnosis, reliability estimate, or validated psychological inference is established. The public version's client-supplied session identifiers and persistent transcripts require privacy and authorization review before use. Stored participant data was not read or copied.

Reproducibility

Reviewed at commit b09a90640dcf on 2026-11-09. Source inspection establishes implementation structure; repository-authored results are not independent measurements. The upstream application and benchmark suite were not rerun during this review.

• src/agent/interview_agent.py
• src/embeddings/axis_scorer.py
• src/db/database.py
• app.py

Next investigation

Resolve the limitations above, preserve the reviewed commit and exact experiment configuration, and reproduce correctness checks before promoting performance or deployment claims.