AI PDF Chatbot

A full-stack AI chatbot that ingests PDF documents, stores embeddings in a vector database, and answers questions using RAG (Retrieval-Augmented Generation). Built with LangChain, LangGraph, Next.js, and Supabase.

Features

• PDF Ingestion — Upload and parse PDFs, store vector embeddings in Supabase
• Smart Query Routing — Automatically decides whether to retrieve documents or answer directly
• Streaming Responses — Real-time SSE-based chat with response chunks
• Source Citations — View which documents were used to generate answers
• Multi-turn Conversation — Message history preserved across turns

Architecture

Frontend (Next.js)  ──>  Backend (LangGraph Server)  ──>  Supabase (Vector Store)
     │                         │
     │  Upload PDFs ──>  Ingestion Graph (embed + store)
     │  Ask questions ──>  Retrieval Graph (route → retrieve → generate)
     │                         │
     └── SSE Stream <──────────┘

• Backend: LangGraph agent graphs for ingestion and retrieval
• Frontend: Next.js/React chat UI with file upload
• Vector Store: Supabase with OpenAI embeddings (text-embedding-3-small)
• LLM: OpenAI GPT-4o-mini (configurable)

Prerequisites

• Node.js v20+
• Yarn
• Supabase project with documents table and match_documents function (setup guide)
• OpenAI API key

Setup

1. Clone the repo:

git clone https://github.com/stevez/pdf-chatbot.git
cd pdf-chatbot

2. Install dependencies:

yarn install

3. Configure environment variables:

Backend (backend/.env):

OPENAI_API_KEY=your-openai-api-key
SUPABASE_URL=your-supabase-url
SUPABASE_SERVICE_ROLE_KEY=your-supabase-service-role-key

Frontend (frontend/.env):

NEXT_PUBLIC_LANGGRAPH_API_URL=http://localhost:2024

Running Locally

Start the backend (LangGraph server on port 2024):

cd backend
yarn langgraph:dev

Start the frontend (Next.js on port 3000):

cd frontend
yarn dev

Open http://localhost:3000, upload a PDF, and start asking questions.

Configuration

• LLM model: frontend/constants/graphConfigs.ts — change queryModel
• Retrieval k value: frontend/constants/graphConfigs.ts — change k
• Prompts: backend/src/retrieval_graph/prompts.ts
• Vector store: backend/src/shared/retrieval.ts

Tech Stack

• LangChain / LangGraph — LLM orchestration
• Next.js — Frontend framework
• Supabase — Vector database
• OpenAI — Embeddings and chat model
• Turborepo — Monorepo management