NG12 Cancer Risk Assessor

A Clinical Decision Support Agent built with Google Vertex AI (Gemini 1.5) to assist in recognizing and referring suspected cancer based on NICE NG12 guidelines.

Features

• Risk Assessment: Analyze patient data (age, symptoms, medical history) against NICE NG12 criteria.
• RAG (Retrieval-Augmented Generation): Answers are strictly grounded in official NICE guidelines with page-level citations.
• Conversational AI: Multi-turn chat interface for exploring cancer referral criteria.
• Patient Integration: Mock integration with a simulated JSON patient dataset.

Tech Stack

• Framework: FastAPI
• LLM: Google Vertex AI (Gemini 1.5 Flash)
• Vector DB: ChromaDB
• Libraries: LangChain, Pydantic, PyPDF

Setup & Installation

Prerequisites

• Python 3.9+
• Node.js & npm
• Google Cloud Project with Vertex AI API enabled
• Application Default Credentials (ADC) for Google Cloud
• (Optional) Docker & Docker Compose

Initial Setup

1. Clone the repository and navigate to the project directory.
2. Create a .env file based on the template:

   GCP_PROJECT_ID=your-project-id
   GCP_LOCATION=us-central1

3. Initialize the backend environment and ingestion:

   python3 -m venv venv
   source venv/bin/activate
   pip install -r requirements.txt
   python ingest_guidelines.py

Running the Application

Option 1: Local Development (Recommended)

1. Backend (FastAPI)

Run this from the root directory:

source venv/bin/activate
python main.py

The backend will be available at http://localhost:8000.

2. Frontend (React + Vite)

Run this from the /frontend directory:

cd frontend
npm install
npm run dev

The frontend will be available at http://localhost:5173.

Option 2: Running with Docker

If you prefer a containerized environment:

docker-compose up --build

Frontend Features

• Patient Assessment: Enter a Patient ID to get an AI-powered cancer risk assessment grounded in NICE NG12 guidelines.
• NG12 Chat: A conversational agent to ask questions about the guidelines with multi-turn support and citations.
• Modern UI: Built with React, Tailwind CSS, and Framer Motion for a premium, responsive experience.

API Endpoints

Assessment

POST /assess

• Body: {"patient_id": "PT-101"}
• Response: Risk assessment with reasoning and citations.

Chat

POST /chat

• Body: {"message": "What symptoms trigger lung cancer referral?", "history": []}
• Response: Grounded explanation with citations.

Implementation Details

• Reasoning Loop: The agent first retrieves patient data, then queries the vector store for relevant criteria, and finally synthesizes the assessment.
• Guardrails: Strict system prompts prevent the agent from providing medical advice outside the scope of the NICE guidelines.
• Citations: All responses include references like [NICE NG12, Page X] for transparency.