Conversational AI Chatbot with Memory

πŸ“˜ Description

This project implements a dynamic, memory-enabled conversational agent that interacts over multiple turns, reasons through ReAct, and streams its thought process using LangGraph and GPT-4o-mini. Built within a Jupyter Notebook, the assistant simulates natural conversation with embedded search capabilities and message-based memory.

The chatbot uses LangChain’s tool-calling logic with Tavily for real-time web answers, scoped thread_id sessions for memory isolation, and a LangGraph agent executor to perform reasoning and retrieval steps transparently.

βΈ»

πŸ”§ Features

  • Agent Executor (LangGraph): Structured loop for reasoning, tool use, and memory recall.
  • Tavily Tool: Live web results fetched automatically through natural language queries.
  • Threaded Memory: Conversation history stored using MemorySaver for scoped sessions.
  • Token Streaming: Displays LLM responses in real time for enhanced UX.
  • Tool Observation: Logs intermediate thoughts and results for explainability.
  • Session ID Control: Allows multiple chats without memory overlap using thread_id.

βΈ»

πŸ’‘ Key Insight

Chatbots become far more powerful when they reason step by step, track what’s been said, and know when to search. This project demonstrates that stateful, tool-augmented chat interfaces are the future of intelligent assistants.

πŸ”— View the source code on GitHub