This repository contains the backend implementation for the Pazago RAG Assignment: Berkshire Hathaway Intelligence, built using the Mastra framework. It utilizes OpenAI's GPT-4o model to create an intelligent RAG agent grounded in Berkshire Hathaway’s annual shareholder letters (2019–2024).
- Name: Harshit Mhatre
- College ID: TU8F2122024
This repository is part of the backend assignment given as part of the Pazago Drive.
- ✅ Mastra Project Initialization
- ✅ Environment Configuration
- ✅ PDF Parsing for Shareholder Letters
- ✅ Embedding Generation via
text-embedding-3-small - ✅ Vector Storage using PostgreSQL +
pgvector
📌 PostgreSQL Table (public.berkshire_embeddings) Screenshot:
- ⚙️ Building the AI Agent
- 💬 Adding memory, citations, and retrieval logic
- 🧠 Integrating Mastra tools for search
| Component | Tool/Framework |
|---|---|
| Framework | Mastra |
| Language | TypeScript |
| Embedding Model | OpenAI text-embedding-3-small |
| LLM | OpenAI GPT-4o |
| Vector DB | PostgreSQL + pgvector |
| PDF Parsing | pdf-parse (Node.js) |
All shareholder letters (2019–2024) are downloaded from this Google Drive folder and parsed for embeddings.
# Clone the repository
git clone https://github.com/xHarshit/Pazago-mastra-app-Harshit.git
cd Pazago-mastra-app-Harshit
# Install dependencies
npm install
# Setup environment
cp .env.example .env
# Fill in OpenAI API key and PostgreSQL credentials
# Parse all PDFs and extract raw text
npx tsx scripts/documentParser.ts
# Generate embeddings and store them in PostgreSQL
npx tsx scripts/documentIngestion.ts
# (In progress) Run the AI Agent
npx tsx berkshire-agent.ts| Phase | Status |
|---|---|
| Mastra Setup | ✅ Complete |
| Document Ingestion | ✅ Complete |
| Vector DB Integration | ✅ Complete |
| AI Agent Implementation | 🔄 In Progress |
| Frontend Chat UI | ⏳ Pending |
Build a production-ready RAG system that:
- Answers financial questions using real-time streaming
- Maintains conversation memory
- Cites exact shareholder letters
- Follows Mastra’s backend architecture