Building and deploying LLMs: A course made by practitioners
Although published a year ago, this Mastering LLMs course remains a good resource for understanding LLM best practices. With teachers like Jeremy Howard and Simon Willison, it offers practical insights from people actually building AI products.
I particularly appreciate the deployment architecture perspectives. For those making technology decisions, understanding deployment complexity matters:
Simple Model Serving: Basic setup with minimal configuration, direct integration with model library (e.g., FastAPI), ideal for proof of concepts and simple applications
Advanced Model Serving: Complex setup involving multiple components and configurations, uses auto-scaling clusters, load balancers, and specialized components, designed for high availability and scalability, suitable for large-scale, production-level applications
Course contents
The course covers these topics:
- Evaluations: Testing frameworks for LLM applications
- RAG (Retrieval-Augmented Generation): Information retrieval systems
- Fine-tuning: LoRA adaptations and QLoRA optimisation
- Building applications: Deployment patterns and serving architectures
- Prompt engineering: Production techniques
Each section includes chapter summaries, notes, and resources. I especially like how they discuss building actual LLM-powered applications and model serving.
The course is free and open. It covers applied topics rather than theory.
