Personal Intelligence App: RAG-based Chatbot

A personal intelligence engine that turns your portfolio and experience into an interactive, context-aware conversational agent. It uses first-person perspective to answer questions about your background, projects, and experience as if it were you.

Tech Stack

• Framework: Next.js 16 (App Router)
• Language: TypeScript
• Styling: Tailwind CSS v4
• AI Models:
  • LLM: Groq (Llama-3.3-70b-versatile)
  • Embeddings: Google Generative AI (gemini-embedding-001)
• Vector Database: DataStax Astra DB
• State Management: Vercel AI SDK

System Architecture

The application follows a RAG (Retrieval-Augmented Generation) pipeline:

How It Works

1. Ingestion (scripts/loadYaml.ts):

   • Recursively walks career_brain/ directory.
   • Parses YAML files (facts, projects, roles).
   • Generates embeddings using Google's gemini-embedding-001.
   • Stores text chunks + vectors in Astra DB.

2. Retrieval (lib/astra.ts):

   • On user query, generates a query embedding.
   • Performs a cosine similarity search in Astra DB.
   • Retrieves the top 5 most relevant context chunks.

3. Generation (app/api/chat/route.ts):

   • Constructs a strict system prompt defining the "Islam Hafez" persona.
   • Injects the retrieved context.
   • Streams the response from Groq's llama-3.3-70b-versatile model.