Meeting Overview

Mentor Meeting — April 17, 2026

The team demonstrated two major achievements:

LLM Research (Aizat) – A systematic comparison of three models (GPT‑4, two open‑source models with 70B and 8B parameters) using a structured prompt and a predefined dialogue. The research was documented in a Jupyter notebook. All models performed well; the main difference was cost per dialogue. The mentor was satisfied with the research artifact.
Frontend + Backend Integration (Ilnar & Timur) – A working demo running via docker-compose showing:
- Case library with 5 pre‑loaded cases
- User login (learner role)
- Case start screen with meta‑information
- Chat UI (ready but not yet connected to LLM)
- Patient summary and notes stubs (partially implemented)

The mentor acknowledged visible progress.

Next priorities:

Connect the LLM to the backend so the chat actually works.
Split the “conversational engine” tasks into basic chat (for MVP) and advanced features (emotion, tone, guardrails) – the latter can be postponed.
Create explicit GitHub issues for LLM integration and assign owners.
Start preparing the final presentation (due May 12, ~3 weeks left).
Consider meeting with real doctors after LLM is integrated, to collect feedback for future iterations.

The mentor also reminded the team to use Pull Requests for visibility and to create a task board (“etazhny bort”) to track who is doing what.

Decision	Rationale
Prioritise basic LLM integration – get a working chat by next week.	MVP needs a demonstrable conversation; emotions and tone can come later.
Postpone real‑time guardrails – only log conversations for offline analysis.	Reduces complexity for the MVP; logs can be analysed later to improve prompts.
Do not meet doctors before LLM is integrated – first get a working prototype, then collect medical feedback.	The team will have something concrete to show and discuss.
Start presentation early – allocate time for iterative improvements.	Only 3 weeks remain until the final presentation (May 12).