Part 07 — Applications
What you actually ship.
The previous parts are the components. This part is about assembling them into the three product shapes that AI-application roles are most often hired to build. Each chapter is written from the failure mode backwards: what does this system look like when it works, and what are the specific ways it breaks.
- Intent Recognition and Routing — classifying what the user wants before deciding which pipeline handles it. Why a small fine-tuned classifier often beats an LLM call on cost and latency, how to handle out-of-scope queries, and how routing mistakes compound.
- NL2SQL (Text-to-SQL) — turning a question into a query against a real schema. Schema linking, the execution-accuracy metric, why valid SQL and correct SQL are different problems, and how to make the system safe to run against production.
- AI-Assisted Coding — repository context, edit formats, verification loops, and what the benchmark scores do and do not measure.
How to read this part: each chapter has a “Using it in a project” section. If you are preparing for an interview that is project-focused — which, for AI-application roles, it usually is — this is the part to over-prepare.
Depends on: Part 05 and Part 06 — each application here assembles a loop, a retrieval pipeline, or both.
Where this leads: Part 08 switches from shipped systems to whiteboard implementation.