RAG API Service (Godspeed)

Overview

A FastAPI-based service for embedding and querying documentation (including code, text, images, PDFs) using a vector database (FAISS) and a Retrieval-Augmented Generation (RAG) agent powered by a multimodal library (CLIP).

Features

• Ingest Git Repository: Clone any public Git repo (and optionally a branch) via /ingest/git and embed .txt, .md, .pdf, .png, .jpg using CLIP.
• Upload File: POST a single file to /ingest/upload (text, PDF, image) and upsert into FAISS.
• Vector Database: Persist embeddings in FAISS (tmp/vector_store.index) with document metadata (tmp/documents.pkl).
• Query API: /query/ accepts a JSON body (question, top_k) and returns a RAG-generated answer plus source list and citations.
• Multimodal Embeddings: Use CLIPModel for both text and images.
• Swagger UI: Interactive API docs at http://<host>:<port>/docs.

---

Prerequisites

• Python 3.9+
• Git CLI
• (Optional) Docker & Docker Compose
• A Perplexity API Key (set PERPLEXITY_API_KEY in .env)

---

Installation

1. Clone this repo

   git clone https://github.com/your-org/assignment.git
   cd assignment
   

2. Create & activate a virtual environment

   python -m venv .venv
   source .venv/bin/activate    # Linux/macOS
   .\.venv\Scripts\activate
   

3. Install dependencies

   pip install --upgrade pip
   pip install -r requirements.txt
   
   

4. Setup environment variables

   cp .env.example .env

Edit .env and set your PERPLEXITY_API_KEY

Running Locally

From project root

   uvicorn app.main:app --host 0.0.0.0 --port 2000

Docker

1. Build

   docker build -t assignment .
   

2. Run

   docker run -d --name assignment -p 2000:2000 \
   -e PERPLEXITY_API_KEY="<your_key>" \
   assignment
   

3. Swagger UI http://localhost:2000/docs

Ingest Endpoints

1. Ingest Git Repo

   POST /ingest/git
   Content-Type: application/json
   
   {
     "repo_url": "https://github.com/your/repo.git",
     "branch":   "main"      # optional, default is default branch
   }
   
   

Response

{ "status": "Ingestion complete", "documents": <count> }

2. Upload File

   POST /ingest/upload
   Content-Type: multipart/form-data
   
   file: <upload your .txt | .md | .pdf | .png | .jpg | .jpeg>

Response

{ "status": "Ingestion complete", "document": "<path>" }

Query Endpoint

POST /query/
Content-Type: application/json

{
 "question": "Your question here",
 "top_k": 3          # optional, default = 3
}

Response

   {
  "response":      "<LLM answer>",
  "sources":       ["path/to/doc1", "path/to/doc2", ...],
  "web_citations": [ ... ]
}

Configuration

# .env
PERPLEXITY_API_KEY=your_perplexity_key
BASE_DIR=tmp/gs_docs
INDEX_PATH=tmp/vector_store.index
DOCUMENTS_PATH=tmp/documents.pkl

Metrics and monitoring

1.Health Check: GET /health → { "status": "ok", "version": "0.1.0" }
2.List Routes: GET /routes → shows all registered endpoints.