🩺 FieldMedic

FieldMedic is an AI-powered clinical decision support system designed to assist frontline healthcare workers by providing evidence-backed patient triage recommendations.

The system combines Retrieval-Augmented Generation (RAG), semantic search, clinical safety rules, and a Large Language Model to produce structured triage decisions.

---

Screenshots

Home

Triage Result

Features

• AI-powered patient triage
• Retrieval-Augmented Generation (RAG)
• Semantic search using Sentence Transformers
• PostgreSQL + pgvector vector database
• FastAPI backend
• Streamlit frontend
• Rule-based clinical safety layer
• Structured JSON outputs
• Knowledge ingestion pipeline
• Evaluation harness for testing

---

System Architecture

                Streamlit UI
                      │
                      ▼
               FastAPI Backend
                      │
        ┌─────────────┴─────────────┐
        ▼                           ▼
 Retrieval Pipeline          Safety Rules
        │                           │
        └─────────────┬─────────────┘
                      ▼
              Gemini LLM Reasoner
                      │
                      ▼
            Structured Triage Output
                      ▲
                      │
       PostgreSQL + pgvector Knowledge Base

---

Tech Stack

• Python
• FastAPI
• Streamlit
• PostgreSQL
• pgvector
• Docker
• Sentence Transformers
• Google Gemini API

---

Project Structure

FieldMedic/
│
├── src/
│   ├── agents/
│   ├── prompts/
│   ├── infra/
│   ├── eval/
│   └── app.py
│
├── ui/
│   └── Home.py
│
├── init/
│
├── Dockerfile
├── compose.yaml
├── requirements.txt
└── README.md

---

Installation

Clone the repository

git clone https://github.com/SandeepKr24/FieldMedic.git
cd FieldMedic

Create a virtual environment

python -m venv .venv

Install dependencies

pip install -r requirements.txt

Create a .env file using .env.example.

---

Running with Docker

docker compose up --build

---

Services

FastAPI

http://localhost:3000/docs

Streamlit

http://localhost:8501

---

API Endpoints

POST /triage

Accepts patient symptoms and vitals and returns a structured triage recommendation.

POST /knowledge

Adds medical knowledge to the vector database.

GET /status

Health check endpoint.

---

Example Response

{
    "triage": "Urgent",
    "care_plan": "Immediate physician assessment and hydration.",
    "icd10": [
        "J18.9"
    ],
    "summary": "Patient presents with fever and productive cough. Evidence suggests possible community-acquired pneumonia.",
    "confidence": 0.92
}

---

Future Improvements

• Images knowledge ingestion
• Multi-provider LLM support
• Authentication
• Explainable AI citations
• Clinical audit logs
• Deployment on cloud infrastructure

---

License

MIT License