RAG Chatbot - AI Knowledge Assistant

(Note: This project will be updated in the future to have an improved and more accessible frontend!)

A Retrieval-Augmented Generation (RAG) application built with Next.js and the Vercel AI SDK. This application allows administrators to ingest PDF documents, converts the text into vector embeddings, and provides a chat interface that answers user questions based on the context of the uploaded data.

Features

• Vector Search: Stores and queries 1536-dimensional embeddings using Neon PostgreSQL with the pgvector extension and HNSW indexing.
• Document Ingestion: Processes uploaded PDFs by extracting text and segmenting it into overlapping chunks using LangChain and the Vercel AI SDK.
• Access Control: Restricts document upload functionality to authorized administrators using Clerk JWT session claims and Next.js edge middleware.
• Streaming Responses: Utilizes the Vercel AI SDK to stream conversational responses back to the user interface.

Tech Stack

• Framework: Next.js (App Router)
• Language: TypeScript
• AI Libraries: Vercel AI SDK, LangChain
• Database: Neon Serverless PostgreSQL
• ORM: Drizzle ORM
• Database Extension: pgvector
• Authentication: Clerk
• Styling: Tailwind CSS, Shadcn UI
• Linting/Formatting: Biome

Getting Started

Prerequisites

You will need Node.js installed on your machine, along with active accounts for Neon (PostgreSQL), Clerk (Authentication), and OpenAI (LLM and Embeddings).

1. Clone the repository

git clone [https://github.com/samiyun/AI-SDK-RAG-Chatbot.git](https://github.com/samiyun/AI-SDK-RAG-Chatbot.git)
cd AI-SDK-RAG-Chatbot

2. Install dependencies

npm install

3. Set up Environment Variables

Create a .env.local file in the root directory and add your configuration keys:

(example)
# Authentication (Clerk)
NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY=your_clerk_publishable_key
CLERK_SECRET_KEY=your_clerk_secret_key

# Database (Neon PostgreSQL)
DATABASE_URL=postgresql://user:password@ep-cool-cloud.region.aws.neon.tech/neondb

# AI Providers
OPENAI_API_KEY=your_openai_api_key

4. Database Setup & Migrations

npx drizzle-kit push

5. Run the Development Server

Open http://localhost:3000 with your browser to interact with the application.

npm run dev

Overview

1. Upload: An authorized administrator uploads a PDF document. The file is parsed and passed to LangChain splitters to divide the text into overlapping chunks.

2. Embedding: The text chunks are sent to OpenAI's embedding model to generate vector representations.

3. Storage: The vectors and their corresponding text chunks are stored in a Neon PostgreSQL database using the pgvector extension.

4. Retrieval: When a user submits a question, the query is converted into a vector. An HNSW-indexed similarity search retrieves the closest matching document chunks from the database.

5. Generation: The retrieved text chunks and the user's prompt are sent to the LLM via the Vercel AI SDK to generate a response based on the provided context.