ametsCS/biorce-mini
An Agentic RAG demo that researches and drafts clinical trial protocols. Built with LangGraph and Streamlit to showcase multi-step AI orchestration, automated reviewing, and UI observability.
What's novel
An Agentic RAG demo that researches and drafts clinical trial protocols. Built with LangGraph and Streamlit to showcase multi-step AI orchestration, automated reviewing, and UI observability.
Code Analysis
11 files read · 3 roundsA Streamlit-powered RAG demo that uses LangGraph to orchestrate a retrieve-research-write-review pipeline for drafting short clinical trial protocol summaries from local text documents, with a bounded revision loop and Langfuse observability.
Strengths
Clean separation of concerns across 4 small source files with a deterministic citation validator as a guard before LLM review, well-structured prompts with strict output formats, and meaningful unit tests for all pure functions. The LangGraph state machine with conditional routing and bounded loop is a solid pattern for agentic workflows.
Weaknesses
The core logic is thin — it's essentially 4 LLM prompt calls wrapped in a state machine with character-based chunking, no retry logic, no semantic chunking, and the revise node is a near-no-op that just passes through the reviewer's suggested draft. The bulk of the repo (154 files) is AI agent/skill configuration and documentation rather than implementation.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
20
Contributors
1
Files
80
Active weeks
2
Repository
Language
Python
Stars
1
Forks
0
License
MIT