CementGPT (FastAPI)

How it works

• Upload: Files are uploaded to data/<session_id>/, split, embedded, and saved as a FAISS index in faiss_index/<session_id>/.
• Chat: Each request loads the FAISS index for the given session_id and answers using RAG.
• Sessions: A simple in-memory history per session on the server (resets on restart). The browser stores session_id in localStorage.

Run locally

1. Install deps

pip install -r requirements.txt

2. Start the server

uvicorn main:app --reload

3. Open the FastAPI Swagger

open http://localhost:8000/docs/

Endpoints

• GET / – Serves the UI.
• GET /health – Health check.
• POST /upload – Form-data file upload. Returns { session_id, indexed }.
• POST /chat – JSON body { session_id, message }. Returns { answer }.

Evaluations 🧪

Run LangSmith evaluations on your RAG system:

# Quick start - run with default settings
python run_evaluations.py

# Run with all evaluators
python run_evaluations.py --evaluator all

# Custom parameters
python run_evaluations.py --evaluator correctness --chunk-size 500 --k 10

Available Evaluators:

• correctness - Custom LLM-as-a-Judge (Gemini 2.5 Pro)
• cot_qa - Chain-of-Thought QA evaluator
• all - Run all evaluators

Notes

• Ensure your API keys/config are set for the ModelLoader to load embeddings/LLM.
• For evaluations, you need LANGSMITH_API_KEY and GOOGLE_API_KEY in your .env file.
• Supported file types: .pdf, .docx, .txt.
• For production, add persistence for chat history and auth; consider cleanup of old session directories.