๐Ÿง  Lexo

LLM Toolkit for RAG, Fine-tuning, and AI Agents

Jupyter Notebooks CPU/GPU Production Ready Open Source

๐ŸŽฏ Project Overview

Lexo is a comprehensive collection of Jupyter notebooks designed for learning and applying Large Language Models (LLMs) in real-world scenarios. Master RAG systems, fine-tuning, AI agents, multimodal processing, and ML benchmarks through hands-on projects.

๐Ÿ“š Each notebook includes:

  • Clear problem statement and real-world context
  • Step-by-step implementation with explanations
  • Required API keys and setup instructions
  • Customizable parameters for experimentation
  • Performance evaluation and insights

๐ŸŽฏ Key Features

๐Ÿง  Core Skills Covered

  • โœจ Retrieval-Augmented Generation (RAG)
  • โœจ Prompt engineering and optimization
  • โœจ AI agents and tool integration
  • โœจ Fine-tuning and QLoRA techniques

๐Ÿค– Models & Technologies

  • โœจ Frontier LLMs (GPT, Claude, DeepSeek)
  • โœจ Open-source models (LLaMA, Mistral)
  • โœจ Speech-to-Text processing
  • โœจ Vector databases and embeddings
๐ŸŽ“

Certification Project

This comprehensive toolkit is the result of my certification from Ed Donner's LLM Engineering Master Course

๐Ÿ”ง Core AI Applications

๐ŸŒ WebPage Summarizer

Beginner
Summarize any URL using OpenAI + LLaMA with Selenium for handling both static and JavaScript-rendered websites.
  • Handles JavaScript websites
  • Markdown-formatted summaries
  • Real-time processing
Selenium BeautifulSoup LLms Ollama

๐Ÿงพ Brochure Generator

Intermediate
Transform websites into AI-crafted brochures for clients, investors, and recruits using intelligent content extraction.
  • Smart content filtering
  • Real-time streaming output
  • Multi-model support
BeautifulSoup LLMs Ollama IPython

๐Ÿ’ก Tech Assistant

Beginner
AI-driven tool that provides concise, structured explanations for technical questions and code snippets.
  • Interactive Q&A
  • Real-time streaming
  • Code explanation
LLMs Ollama IPython

๐Ÿค– TriBot Debate

Intermediate
Three-bot chat system with GPT (polite & humorous), Claude (argumentative & snarky), and DeepSeek (logical & analytical).
  • Distinct personalities
  • Multi-model integration
  • Customizable prompts
OpenAI Anthropic DeepSeek IPython

๐ŸŒค๏ธ WeatherMate AI Agent

Intermediate
Conversational AI agent that analyzes real-time weather conditions and suggests activities and events based on location.
  • Real-time weather data
  • Event recommendations
  • External API integration
LLMs Tools REST APIs Gradio

๐Ÿ“ Advanced Workflows

๐Ÿ“ Meeting Minutes Assistant

Intermediate
Generate structured meeting minutes from audio recordings using Speech-to-Text (Whisper) and Large Language Models.
  • Audio transcription
  • Structured output
  • Real-time streaming
Whisper LLaMA 3.1 Gradio HuggingFace GPU

๐Ÿงช Synthetic Data Generator

Intermediate
Generate realistic synthetic datasets for tabular, text, and time-series data using multiple LLM providers.
  • Multiple data types
  • JSON and CSV output
  • Multi-model support
OpenAI Anthropic Google Cloud Platform Gradio

๐Ÿง  RAG QA Assistant

Intermediate
Internal expert knowledge assistant using Retrieval-Augmented Generation (RAG) for fast, accurate answers to internal queries.
  • Document loading (PDF, text, markdown)
  • ChromaDB vector store
  • Conversation history
  • Source attribution
LangChain ChromaDB LLMs Gradio

๐Ÿ”ฌ ML & Fine-tuning Pipeline

Complete ML pipeline from data to deployment with comprehensive benchmarking and evaluation. GPU required for optimal performance.

๐Ÿ“Š Data Curation

Aggregate, clean, analyze, and balance datasets for price prediction tasks.

โš”๏ธ Traditional ML vs LLMs

Compare traditional ML models against frontier LLMs for performance benchmarking.

๐Ÿง  E5 Embeddings & RAG

Test contextual embeddings and retrieval-augmented generation approaches.

๐Ÿ”ง Fine-tuning GPT-4o Mini

Fine-tune frontier models and compare before/after performance.

๐Ÿฆ™ LLaMA 3.1 Evaluation

Evaluate quantized LLaMA 3.1 8B model performance.

โš™๏ธ QLoRA Fine-tuning

Fine-tune LLaMA 3.1 using QLoRA with hyperparameter optimization.

๐Ÿงช Model Evaluation

Comprehensive evaluation and performance comparison across all models.

๐Ÿ† Leaderboard

Final rankings and insights across ML, embeddings, RAG, and fine-tuned models.

๐Ÿ† Capstone Project

๐Ÿท๏ธ Snapr - AI Deal Finder

Advanced
Capstone Project: A comprehensive AI system that scans online product listings, predicts their value using an ensemble of models, and alerts users to great deals. This project integrates the ensemble model (fine-tuned LLaMA, XGBoost, and GPT-4o Mini + RAG) with cloud deployment (Modal/GCP) for production use.
  • Multi-model ensemble integration
  • Real-time price prediction
  • Cloud deployment with Modal
  • Scalable production infrastructure
  • Deal alert system
LLaMA 3.1 Fine tuned with QLoRA XGBoost RAG ChromaDB/AWS S3 Modal Docker HuggingFace Modal GPU Google Cloud Platform

๐ŸŒŸ Explore More Projects

Inspired by these AI solutions? Explore more machine learning and artificial intelligence projects showcasing advanced techniques and real-world applications.