This repository provides a free, hands-on course for building a production-ready LLM and RAG system, specifically an "LLM Twin" that mimics a user's writing style. It targets ML/AI engineers, data engineers, data scientists, and software engineers looking to understand and implement end-to-end LLM system engineering with LLMOps best practices. The benefit is learning to deploy a functional AI replica from data collection to inference.

How It Works

The project is architected as four Python microservices: data collection, feature engineering, training, and inference. It employs a streaming pipeline using Bytewax for real-time data processing, vector databases (Qdrant, Redis) for RAG, and AWS SageMaker for LLM fine-tuning and deployment. Key LLMOps practices like experiment tracking (Comet ML), model versioning (Hugging Face), and prompt monitoring (Opik) are integrated. Bonus lessons cover optimizations using Superlinked.

Quick Start & Requirements

• Install/Run: Follow the INSTALL_AND_USAGE document for step-by-step instructions.
• Prerequisites: Basic Python and Machine Learning knowledge.
• Hardware: Any modern laptop/workstation; fine-tuning and inference are cloud-based (AWS SageMaker).
• Dependencies: AWS account, OpenAI API key (minimal cost).
• Resources: Free tier tools are used, with estimated costs for AWS and OpenAI being minimal (<$10).
• Docs: INSTALL_AND_USAGE

Highlighted Details

• End-to-end LLM & RAG system implementation.
• Integrates 4 serverless tools: Comet ML, Qdrant, AWS SageMaker, Opik.
• Demonstrates Change Data Capture (CDC) and real-time streaming with Bytewax.
• Includes advanced RAG techniques and prompt monitoring.
• Bonus lessons on optimizing RAG with Superlinked.

Maintenance & Community

• Open-source course with contributions welcomed.
• Sponsors include Comet, Opik, Bytewax, Qdrant, and Superlinked.
• Updates via Decoding ML newsletter.

Licensing & Compatibility

• MIT License.
• Permissive for commercial use, university projects, and personal projects, provided attribution is given.

Limitations & Caveats

The course focuses on engineering practices and system implementation, not theoretical model optimization or research. While free to use, it requires an AWS account and an OpenAI API key, incurring minimal costs for cloud resources and API calls.