🧠 Booprogram Agent

A lightweight AI-powered document ingestion and querying system built with React and OpenAI. Upload documents (PDF, DOCX), process them into chunks, and query them using embeddings + LLM reasoning.

🚀 Features

• 📄 Upload and parse PDF and DOCX files
• 🧩 Automatic chunking & embedding of document content
• 🔍 Semantic search over uploaded documents
• 💬 Ask questions and get AI-generated answers based on your data
• 🗂️ Document management (list + delete)

📦 Installation

npm install booprogram-agent

⚙️ Setup

1. Add your OpenAI API key

Create a .env file in your project root:

OPENAI_API_KEY=your_openai_api_key_here

2. Basic Usage

import { AgentDashboard } from "booprogram-agent";

export default function Page() {
  return <AgentDashboard apiKey={process.env.OPENAI_API_KEY!} />;
}

🧠 How It Works

1. Upload Document

   • Files are parsed (pdf-parse / docx)
   • Text is extracted

2. Processing

   • Text is split into chunks
   • Each chunk is embedded using OpenAI embeddings

3. Storage

   • Chunks + embeddings are stored locally (or your backend)

4. Querying

   • User question → embedded
   • Similar chunks are retrieved
   • Context is sent to LLM → answer generated

🛠️ Supported File Types

• PDF
• DOCX

📁 Project Structure (Simplified)

/parsers
  parsePDF.ts
  parseDocx.ts

/knowledge
  ingestDocument.ts
  getDocuments.ts
  deleteDocument.ts

/components
  AgentDashboard.tsx

⚡ Notes on Token Usage

• Documents are not sent entirely to OpenAI

• Only relevant chunks are retrieved and used

• This keeps:

  • 💰 costs low
  • ⚡ responses fast

🧪 Future Improvements

• Vector DB integration (Pinecone / Weaviate)
• Streaming responses
• Better chunking strategies
• UI improvements (drag & drop, previews)
• Multi-document context reasoning

🧑‍💻 Development

npm run dev

📄 License

MIT

🤝 Contributing

PRs and suggestions are welcome.

🧩 Author

Built by Lazar Velickovic