StudyVault

StudyVault is a personal AI study assistant that lets you build a private knowledge library from PDF documents and YouTube videos, then ask natural language questions answered exclusively from your own uploaded material. It uses a multi-agent RAG (Retrieval-Augmented Generation) pipeline to semantically chunk and index content, retrieve the most relevant passages at query time, and generate cited answers via an LLM.

---

Agents

Agent	Purpose
YouTubeAgent	Extracts transcripts from YouTube URLs with a 3-layer fallback: English → any available language → MCP subprocess
ChunkingAgent	Splits raw text into semantically coherent chunks using SemanticChunker (95th-percentile cosine boundaries), falling back to RecursiveCharacterTextSplitter
IndexingAgent	Embeds chunks with OpenAI text-embedding-3-small and upserts them as vector points to Qdrant, then writes point IDs and chunk count back to PostgreSQL
RAGOrchestrator	Wires all agents together via LangChain LCEL chains; handles ingestion pipelines for PDFs and YouTube, filtered vector search, prompt construction, and LLM response generation

---

How to Run

Prerequisites

• Docker Desktop installed and running
• An OpenAI API key (required for embeddings and LLM fallback)

Setup

1. Create the backend environment file:

   code/webapp/backend/.env.docker
   

   Minimum required contents:

   OPENAI_API_KEY=sk-...
   DATABASE_URL=postgresql://postgres:postgres@db:5432/studyvault
   SECRET_KEY=your-secret-key-here
   # Optional — if omitted, GPT-4o-mini is used instead
   GEMINI_API_KEY=...

2. Start all services from the code/ directory:

   cd code
   docker-compose up --build

3. Open the app in your browser:

   • Frontend: http://localhost:3000
   • Backend API: http://localhost:8000
   • Qdrant Dashboard: http://localhost:6333/dashboard

Stopping

docker-compose down

Add -v to also delete the database and vector store volumes:

docker-compose down -v