Meeting Overview

1. Meeting Summary

The team presented progress on backend case management and frontend case library. Key discussions focused on:

  • Demo readiness – The mentor requested a working demonstration of the chat interface and case selection.
  • LLM experiments – The team has conducted initial tests with different models (DeepSeek, etc.) but lacks a structured artifact (e.g., Jupyter notebook) showing prompts, responses, and evaluation.
  • API key blocker – The team is waiting for an API key to integrate a production LLM. The mentor is willing to provide a key only after seeing a clear research artifact that proves the chosen approach works.
  • Pull request visibility – The mentor emphasized using open Pull Requests (not just commit history) to make individual work visible to other mentors and stakeholders.
  • Russian language requirement – The final dialogue should be in Russian, though experiments can be in English.

2. Key Technical Status

Backend & Frontend

  • Backend: Case management endpoints completed, seed script available to populate database with synthetic cases.
  • Frontend: Case library view (list of cases) is functional. Chat UI is under development.
  • The team uses OpenAPI spec generation → code generation for frontend requests to ensure contract consistency.

Case Data Structure

  • The team has defined a JSON structure for clinical cases (persona, symptoms, gold standard, etc.).
  • A sample case has been created and tested manually with LLM.

LLM Experiments (Aizat)

  • Tests performed with DeepSeek, Chinese models (produce thinking in parentheses), and others.
  • No Jupyter notebook or systematic evaluation artifact exists yet.
  • The team believes the basic conversational engine can work, but needs API key to integrate into the backend.

3. Mentor’s Core Feedback

On Demonstrating Progress

“If you have done research but have no artifact, you might as well have drunk coffee and done nothing.”

  • Every research activity must produce a tangible artifact (notebook, script, documented test results).
  • The artifact should allow someone else to reproduce the results without spending the same amount of time.
  • For LLM experiments, a Jupyter notebook is the preferred format – showing prompts, model responses, and evaluation against expected answers.

On the “Conversational Virtual Patient Engine” (EPIC-03)

  • The mentor is not convinced that the engine works because there is no visible evidence (no chat UI, no CLI demo, no notebook).
  • Minimum requirement: Show a working conversation – even a simple script that asks a few questions and receives plausible answers.
  • If the team shows a working conversation (even with a free/local model) and a clear evaluation of why the answers are acceptable, the mentor will provide an API key.

On API Key & Proxy

  • The mentor can provide a key but warned that direct Google API usage may be blocked. He suggested setting up a proxy endpoint (e.g., on a separate server) to avoid rate limits / regional blocks.
  • The team can also use free models (DeepSeek, Gemini free tier) for initial prototyping.

On Pull Requests vs Commit History

  • Pull Requests are preferred because they clearly show who is working on what, and mentors can see open PRs at a glance.
  • Commit history is less visible. The team should adopt a workflow where every change starts with a branch and a PR (even if not immediately merged).

On Language

  • The final product should support Russian (target audience is Russian-speaking medical students).
  • Experiments can be in English, but the team must eventually test with Russian prompts and responses.

Table of contents

This site uses Just the Docs, a documentation theme for Jekyll.