CLINICAL REVISOR MCP

• 

  User-friendly Streamlit dashboard for real-time medical auditing and report generation.

• 

  Backend execution logs showing the autonomous agent workflow, node transitions, and LLMOps tracking.

🛡️ Inspiration

The healthcare industry faces a massive challenge in clinical safety auditing. Manual review of patient records and lab results is slow and prone to human error. I wanted to build an autonomous system that could act as a second pair of eyes for doctors, ensuring no clinical rule is overlooked.

🤖 What it does

Clinical Revisor MCP is a multi-modal agentic system that:

• Analyzes Complex Data: Processes PDF patient records and JPG lab results using Gemini Vision.
• Self-Corrects: Features a "Critic" node that audits drafts against a master_rules database, triggering recursive rewrites if safety criteria aren't met.
• Real-time Research: Uses Tavily Search to cross-reference drug interactions and medical guidelines.
• LLMOps: Tracks token usage and API costs in real-time for full transparency.

🛠️ How I built it

The system is architected as a Directed Acyclic Graph (DAG) using LangGraph. I utilized Gemini 3.1 Flash Lite for its speed and multi-modal capabilities, and Tavily for agentic web-retrieval.

🧠 Challenges I ran into

Managing the state between different nodes (Vision, Research, and Critic) required careful design. Ensuring the agent stayed within clinical safety bounds was the top priority.

🏆 Accomplishments that I'm proud of

I successfully built a system that not only analyzes data but actively "thinks" and "revises" its own work until it reaches a safe standard.

What we learned:

Building this project taught us the power of Agentic RAG and the necessity of a 'human-in-the-loop' philosophy via the Critic Node. We learned how to efficiently handle multi-modal inputs (vision + text) and the importance of real-time cost and token tracking in LLMOps to keep the system sustainable.

What's next for CLINICAL REVISOR MCP:

The next step is to integrate full FHIR standard support for direct EHR connectivity. We also plan to enhance the evaluation layer with more granular clinical metrics and expand the vision capabilities to include more complex medical imaging like MRI and CT scans.

Built With

• agentic-rag
• dotenv
• gemini
• langchain
• langgraph
• llmops
• pillow
• pypdf
• python
• tavily