Build AI-Powered Apps

A project-based course by Mosh Hamedani on building LLM applications: from the fundamentals of language models (tokens, model choice, model settings) through a modern full-stack setup with Bun, TailwindCSS and shadcn/ui to two end-to-end projects — a chatbot with a streaming UI and a review summariser with MySQL, Prisma and TanStack Query. It adds a dedicated module on prompt engineering and the use of open-source models via Hugging Face and Ollama. Refactoring runs through the whole course: repository, service and controller are extracted step by step. Fully completed (all lessons).

Contents & competencies

• Language model fundamentals: what LLMs can do, understanding and counting tokens, choosing the right model, model settings and API calls
• Modern full-stack setup: Bun, separate frontend and backend, OpenAI API key handling, TailwindCSS, shadcn/ui, Prettier and pre-commit checks with Husky
• Chatbot backend: chat API, conversation state management, input validation and error handling — then refactored into repository, service, controller and routes
• Chatbot frontend: chat UI, state management, markdown rendering, typing indicator, auto-scrolling and error handling, broken down into separate components
• Prompt engineering: how a good prompt is built, supplying context, steering the output format, giving examples, reducing hallucinations and improving prompts iteratively
• Review summariser: MySQL and Prisma with schema, migrations and realistic test data, an API to fetch and summarise reviews, storing and regenerating the summary, edge cases
• Frontend with TanStack Query: displaying reviews and star ratings, loading skeletons, error handling, mutations and an extracted API layer
• Open-source models: finding and calling models on Hugging Face, choosing the right model for the task, running models locally with Ollama

Syllabus

The course page states 7 hours and 120 lessons across 8 sections but lists individual lessons only for “Getting Started”. The outline here is therefore taken from the course player and shows all 109 lessons that were included at completion in December 2025 — the course has grown since. The Code with Mosh certificate itself states no scope. The numbering of sub-lessons (2.1, 4.3 …) has been removed in favour of continuous counting.

Topics