Farzad-R/LLM-Zero-to-Hundred

This repository contains different LLM chatbot projects (RAG, LLM agents, etc.) and well-known techniques for training and fine tuning LLMs.

What it solves

This repository serves as a comprehensive collection of practical implementations and tutorials for building advanced LLM applications. It bridges the gap between basic chatbot development and production-ready systems by demonstrating various RAG (Retrieval Augmented Generation) techniques, fine-tuning pipelines, and multimodal agentic workflows.

How it works

The repository is organized into several standalone projects, each focusing on a specific AI capability:

  • RAG Systems: Implements multiple retrieval strategies using LangChain and LlamaIndex, including chatbots that can query pre-processed documents, real-time uploads, or specific websites.
  • Agentic Workflows: Uses LangGraph to build chatbots capable of autonomous decision-making, function calling, and internet search via DuckDuckGo.
  • Multimodal Capabilities: Integrates Stable Diffusion for image generation and LLava for image understanding, alongside text and voice interaction.
  • Fine-tuning: Provides a pipeline for processing raw data and fine-tuning multiple LLMs to create a domain-specific chatbot.
  • Open-Source Deployment: Demonstrates on-premise deployment using Google Gemma7B and BAAI embedding models.

Who it’s for

Developers and AI practitioners who want to move from basic LLM prompts to building complex, multimodal, and retrieval-augmented applications with real-world architectural patterns (like microservices and Docker).

Highlights

  • Diverse RAG Implementations: Comparison of five different RAG techniques across LangChain and LlamaIndex.
  • Multimodal Integration: A single chatbot (HUMAIN) that handles text, voice, images, and web search.
  • Production-Style Design: Includes examples of microservice architecture using Docker Compose and visual monitoring with Weave Scope.
  • End-to-End Fine-tuning: A complete pipeline from raw data processing to model evaluation and deployment.