📄 RAG-Based PDF Assistant

https://documentchat4u.streamlit.app/ A production-ready Retrieval-Augmented Generation (RAG) pipeline for querying PDF documents using natural language. Built with FastAPI, Inngest, Streamlit, and Qdrant — fully deployable on Render + Streamlit Cloud with no OpenAI dependency required.

---

🏗️ Architecture

Streamlit Cloud (UI)
        ↓
Render FastAPI (Backend)
        ↓
Inngest Cloud (Workflow Orchestration)
        ↓
Qdrant Cloud (Vector Database)

PDF Upload Flow

Streamlit → POST /upload → Save PDF on Render
         → Trigger rag/ingest_pdf via Inngest
         → Load → Chunk → Embed → Store in Qdrant

Query Flow

Streamlit → POST /query → Trigger rag/query_pdf_ai via Inngest
         → Embed question → Search Qdrant
         → Return extractive answer → Display in UI

---

✨ Features

• Upload and ingest PDF documents via a modern Streamlit dashboard
• Semantic search using sentence-transformers (no OpenAI required)
• Extractive answer mode — returns retrieved context directly
• Inngest-powered durable workflow orchestration with retries and throttling
• Qdrant Cloud vector store with configurable dimensions
• Secure backend proxy for Inngest polling (no secrets exposed in the frontend)
• Deterministic chunk UUIDs to avoid duplicate vectors on re-ingestion
• OpenAI support available as an optional drop-in for embeddings and answer generation

---

🧰 Tech Stack

Layer	Technology
Frontend	Streamlit
Backend API	FastAPI + Uvicorn
Workflow Engine	Inngest
Vector Database	Qdrant Cloud
Embeddings	sentence-transformers/all-MiniLM-L6-v2 (or OpenAI)
PDF Parsing	LlamaIndex (PDFReader, SentenceSplitter)
Deployment	Render (backend), Streamlit Cloud (frontend)

---

🚀 Getting Started

Prerequisites

• Python 3.11+
• uv package manager
• Docker (for local Qdrant)
• Node.js (for local Inngest Dev Server)

1. Clone the Repository

git clone https://github.com/your-username/RAG-based-agent.git
cd RAG-based-agent

2. Install Dependencies

uv sync --frozen

3. Configure Environment

Copy the example env file and fill in your values:

cp .env.example .env

# Inngest
INNGEST_EVENT_KEY=your_event_key
INNGEST_SIGNING_KEY=your_signing_key

# Qdrant
QDRANT_URL=https://your-cluster.qdrant.io
QDRANT_API_KEY=your_qdrant_api_key
QDRANT_COLLECTION=docs_minilm
QDRANT_DIM=384

# Embeddings
EMBED_PROVIDER=sentence-transformers
SENTENCE_TRANSFORMER_MODEL=sentence-transformers/all-MiniLM-L6-v2
EMBED_DIM=384

# Answer generation
ANSWER_PROVIDER=extractive

# Optional: OpenAI
# OPENAI_API_KEY=sk-...

4. Start Local Qdrant (Docker)

docker run -d --name qdrantRAGdb \
  -p 6333:6333 \
  -v "$(pwd)/qdrant_storage:/qdrant/storage" \
  qdrant/qdrant

5. Start the Inngest Dev Server

npx --yes --ignore-scripts=false inngest-cli@latest dev \
  -u http://127.0.0.1:8000/api/inngest --no-discovery

6. Start the FastAPI Backend

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

7. Start the Streamlit Frontend

uv run streamlit run streamlit_app.py

---

📡 API Endpoints

Method	Endpoint	Description
GET	/health	Health check
POST	/upload	Upload a PDF and trigger ingestion
POST	/query	Ask a question over ingested documents
GET	/runs/{event_id}	Poll Inngest run status (server-side proxy)
POST	/api/inngest	Inngest webhook handler

Example: Query

curl -X POST https://your-backend.onrender.com/query \
  -H "Content-Type: application/json" \
  -d '{"question": "What is the main topic?", "top_k": 5}'

---

☁️ Deployment

Render (FastAPI Backend)

Setting	Value
Build Command	pip install uv && uv sync --frozen
Start Command	uv run uvicorn main:app --host 0.0.0.0 --port $PORT

Required environment variables on Render:

INNGEST_EVENT_KEY
INNGEST_SIGNING_KEY
QDRANT_URL
QDRANT_API_KEY
QDRANT_COLLECTION=docs_minilm
QDRANT_DIM=384
EMBED_PROVIDER=sentence-transformers
SENTENCE_TRANSFORMER_MODEL=sentence-transformers/all-MiniLM-L6-v2
EMBED_DIM=384
ANSWER_PROVIDER=extractive

Streamlit Cloud

Add the following to your Streamlit secrets:

FASTAPI_URL = "https://your-backend.onrender.com"
INNGEST_UI_URL = "https://app.inngest.com"

---

🔑 Inngest Cloud Setup

1. Create an account at inngest.com
2. Go to Production environment → Event Keys → create a new event key
3. Go to Signing Key section → copy your signing key
4. Add both to your Render environment variables
5. Register your deployed endpoint in Inngest Cloud:

   https://your-backend.onrender.com/api/inngest
   

---

🧪 Embedding Modes

Mode	EMBED_PROVIDER	Requires
Open-source (default)	sentence-transformers	Nothing — runs locally
OpenAI	openai	OPENAI_API_KEY with credits
Fake (testing only)	set USE_FAKE_EMBEDDINGS=1	Nothing

⚠️ Always use the same embedding mode for ingestion and querying. Mixing modes produces invalid results.

---

📁 Project Structure

RAG-based-agent/
├── main.py              # FastAPI app + Inngest functions
├── data_loader.py       # PDF loading, chunking, embedding
├── vector_db.py         # Qdrant client wrapper
├── custom_types.py      # Shared type definitions
├── streamlit_app.py     # Streamlit frontend
├── .env.example         # Environment variable template
├── .gitignore
└── pyproject.toml       # uv/pip dependencies

---

🔒 Security Notes

• Never commit .env to Git — it is listed in .gitignore
• The /runs/{event_id} endpoint proxies Inngest polling server-side so your INNGEST_SIGNING_KEY is never exposed to the browser or Streamlit Cloud
• Rotate any API keys that were accidentally exposed

---

📄 License

MIT License. See LICENSE for details.

---

🙏 Acknowledgements

• LlamaIndex for PDF parsing and chunking
• Qdrant for vector storage
• Inngest for durable workflow orchestration
• Sentence Transformers for open-source embeddings