Capstone Projects

A capstone project is a complete, guided build that pulls a whole course together. Instead of one technique in isolation, you build a working product with every part that real users depend on: the model calls, the data model, the validation, the evaluation suite, the cost tracking, and the plan for shipping it safely. These six projects are where the patterns from the earlier modules stop being separate lessons and start being one way of working.

Capstones matter because the hard part of AI products is rarely the first prompt. It's the seams: retrieval that misses an exact error code, an extraction that is schema-valid but wrong, a cost line that grows faster than revenue, a rollout nobody can reverse. Each project here is designed around those seams, with real, type-checked TypeScript and the trade-offs explained, so you finish with a product you understand well enough to change.

The module covers six builds. You'll add AI ticket triage to a help-desk SaaS with classification, summaries, drafted replies, evals, and a staged rollout. You'll build a cited question-answering app over a document collection with hybrid search, reranking, and permission filters. You'll turn receipt photos, PDFs, and voice notes into validated expenses with a review screen. You'll build an internal analytics assistant that answers questions over company data with safe queries, charts, and an audit log. You'll ship a product-description generator with brand voice controls, model-graded quality checks, batch generation, and usage-based billing. And on the no-code track, you'll launch a lead-qualification assistant without writing code.

Each capstone links back to the lessons it relies on rather than repeating them, so you can work through the builds in any order and jump back for depth when you need it. The module ends with guidance on keeping your skills current as models change and where to go next, including building agents.