Overview
This tutorial guides you through building a fully local AI assistant with memory capabilities, similar to ChatGPT but without subscription fees. Learn to create a wellness coach chatbot that remembers user information using Gemma 3 and Qwen models via Ollama, combined with LangChain, LangGraph, and SQLite for persistent storage. The 37-minute video covers the complete development process from architecture design to implementation, including vector search memory, conversation history persistence, custom tool creation, and a Streamlit user interface. Perfect for developers wanting to build practical AI applications without cloud dependencies. The tutorial breaks down each component systematically, starting with project setup and configuration, moving through model initialization, database connections with sqlite-vec, chatbot workflow implementation, memory management, tool calling capabilities, and finally the user interface development, concluding with a live demonstration of the finished application.
Syllabus
00:00 - Welcome
00:53 - App architecture
01:57 - Full-text tutorial and source code on MLExpert.io
02:27 - Gemma 3 on Ollama
03:14 - Project overview, config and dependencies
05:55 - Model LLM and embeddings initialization
06:24 - Database SQLite connection with sqlite-vec
07:42 - Chatbot workflow
20:38 - Memory management
25:38 - Custom tool and tool calling
28:42 - UI with Streamlit
33:30 - App demo
35:44 - Conclusion
Taught by
Venelin Valkov