MedRec Sentinel is an audit-first medication reconciliation system designed to streamline discharge workflows. It leverages Google MedGemma as a core extraction engine to identify discrepancies, validate outputs against deterministic safety rules, and generate clinician-ready draft notes.

Developed as part of the MedGemma Impact Challenge, the project focuses on agentic workflows that bridge LLM capabilities with traditional clinical safety checks to ensure reliability in high-stakes medical environments.

• Language: Python 3.10+
• LLM Engine: Hugging Face Transformers & Google MedGemma
• Validation: Pydantic
• UI Framework: Gradio
• Testing: Pytest
• Quantization: BitsAndBytes (4-bit support)

• 🧬 MedGemma-Powered Extraction: High-fidelity clinical data extraction from unstructured discharge notes.
• 🤖 Agentic Workflow: Multi-step pipeline that validates model outputs and runs safety checks.
• 📝 Clinician-Facing Drafts: Automatically generates pharmacist-ready notes with critical verification questions.
• 🛡️ Deterministic Safety Engine: Rules-based engine to catch high-risk medication errors before human review.
• 📊 Comprehensive Evaluation: Built-in evaluation suite to measure extraction accuracy across clinical cases.

• Python 3.10+
• (Optional) CUDA-capable GPU for running MedGemma locally

1. Clone the repository:

   git clone https://github.com/roivroberto/medrec-sentinel.git
   cd medrec-sentinel

2. Set up a virtual environment and install dependencies:

   python3 -m venv .venv
   source .venv/bin/activate
   pip install -r requirements.txt
   pip install -r requirements-dev.txt

To run the baseline evaluation (no model required):

python3 -m medrec_sentinel.eval.run_eval --data data/synth/cases.jsonl --mode baseline

To launch the Gradio demo:

python3 demo/gradio_app.py

This project is licensed under the MIT License.