data/github/personal_ai_agent/readme.md

Personal AI Digital Twin

Problem

Building a professional presence that works 24/7 is hard when you're also building. Most personal websites are static — they don't engage visitors, can't answer questions, and miss potential clients or collaborators entirely.

Solution

A deployed AI agent that speaks as Arun, answers questions about his background, skills, and projects, and captures lead contact details automatically — all without Arun needing to be present. The system runs on Render, notifies Arun on Telegram in real time, and degrades gracefully if any one LLM model hits a rate limit.

Key Features

• Persona context injection: System prompt is built from a summary.txt file containing Arun's professional background — the agent speaks with Arun's voice, not a generic assistant's
• Tool-calling agent loop: Runs a full agentic chat loop with up to 3 iterations per message, allowing the agent to call tools (lead capture, unknown question logging) before generating the final response
• Multi-model fallback chain: Tries llama-3.3-70b-versatile → qwen3-32b → llama-4-scout → llama-3.1-8b-instant in sequence — if any model hits a rate limit, the next one takes over automatically
• Lead capture: Detects hiring/collaboration intent and calls a tool to save lead info locally and push a Telegram notification to Arun's phone
• Session memory: In-memory conversation history per session, trimmed to the last 6 turns to keep token usage under control
• Real-time Telegram logging: Every chat message and AI response is forwarded to Arun via Telegram bot
• Live deployment: Hosted on Render with dynamic port binding via PORT environment variable

Live URL

https://personal-ai-agent-96aq.onrender.com