TransitGuardRAG

RAG-powered safety chatbot for Chicago transit.

Backend API for the TransitGuard platform. Uses Pinecone vector search and Claude Haiku 3 to answer natural language questions about CTA safety incidents, station risk levels, and real-time crime data.

Example Queries

• "What are the stations near me?"
• "Total number of crimes today"
• "Safest line in the last 7 days"
• "Total number of traffic accidents today"

Architecture

Question → SentenceTransformer (all-MiniLM-L6-v2) → Embedding
                                                      ↓
                                              Pinecone Vector Search
                                                      ↓
                                              Top-K Context Chunks
                                                      ↓
                                              Claude Haiku 3 → Answer

---

Features

• /query endpoint: Accepts a POST request with an embedding vector and question, returns a generated answer and sources.
• /health endpoint: Simple health check for the API.
• Special queries: Crime totals, traffic accidents, safest lines, nearby stations
• Python client script: Converts a question to an embedding and queries the API (local or Railway).
• Dockerized & Railway-ready: Easily build and run the API in a container or deploy to Railway.

---

Project Structure

TransitGuardRAG/
│
├── main.py                   # FastAPI app for Pinecone + Claude querying
├── query_pinecone.py         # Python script to embed a question and query local API
├── railway_query_pinecone.py # Python script to query deployed Railway API
├── requirements.txt
├── Dockerfile
├── start.sh                  # Entrypoint for Railway
├── README.md
└── .env                      # Environment variables (not committed)

---

Quick Start

1. Clone the Repository

git clone https://github.com/foxintheloop/TransitGuardRAG.git
cd TransitGuardRAG

2. Install Dependencies

pip install -r requirements.txt

3. Environment Variables

Create a .env file in the root directory:

PINECONE_API_KEY=your-pinecone-api-key
CLAUDE_API_KEY=your-anthropic-claude-api-key

4. Run Locally

uvicorn main:app --reload --host 0.0.0.0 --port 8000

---

Docker Deployment

docker build -t transitguardrag .
docker run --env-file .env -p 8000:8000 transitguardrag

---

Railway Deployment

1. Connect your repo to Railway
2. Set environment variables (PINECONE_API_KEY, CLAUDE_API_KEY) in the Railway dashboard
3. Deploy — Dockerfile and start.sh are pre-configured

---

API Reference

POST /query

Query the Pinecone index and generate an answer with Claude.

Request:

{
  "embedding": [0.1, 0.2, 0.3, ...],
  "top_k": 5,
  "question": "What are the stations near me?"
}

Response:

{
  "answer": "The stations near your current location are: ...",
  "sources": [ ... ]
}

GET /health

Health check. Returns { "status": "healthy" }.

---

Python Client Example

import requests
from sentence_transformers import SentenceTransformer

question = "What are the stations near me?"
model = SentenceTransformer('all-MiniLM-L6-v2')
embedding = model.encode(question).tolist()

response = requests.post(
    "https://your-railway-url.up.railway.app/query",
    json={"embedding": embedding, "top_k": 5, "question": question}
)
print(response.json())

---

Stack

FastAPI Pinecone Claude API SentenceTransformers Docker Railway

---

Related

• TransitGuard — Main project overview
• transitguard-dashboard — Web dashboard
• transitguard-app — Mobile app

---

License

MIT