Overview

This project is a hands-on implementation of a tool-using AI agent built from scratch. The goal was to understand how modern AI agents work internally, avoiding reliance on frameworks like LangChain.

How It Works

The agent follows a simple execution loop (think → act → observe):

1. User Input
2. LLM Decision: Determines whether a tool is needed
3. Tool Execution: Executed via a backend function
4. Observation: Result is passed back to LLM
5. Final Output: The agent generates a final response based on tool data

Features & Integrations

• Tool-based Architecture: Dynamic tool selection via LLM
• Multiple APIs: Country info, crypto prices, weather data, and geolocation.
• Multi-step Execution: The agent handles multi-step interactions seamlessly.

Key Takeaways

This project deeply solidified my understanding of how LLMs interact with tools, how agent loops are implemented from scratch, and the actual utility of orchestration frameworks versus raw API calls.