AI Full Stack

What sits between the endpoint and the user: token streaming, retrieval, evaluation, and routing across models.

For: AI application engineers, and anyone proving that a model swap did not break a product.

Included in: Superstar FDE. Progress made here counts there too.

The stages, their chapters, and what “done” means for each are in path.tsv.

practice/bin/ol learn ai-full-stack          # stages and progress
practice/bin/ol learn ai-full-stack next     # read the next unfinished stage

Stages

Generated from path.tsv. Track progress locally with practice/bin/ol learn.

Stage	Read	Done when
1	Streaming and SSE Streaming & SSE	You stream tokens from an OpenAI-compatible endpoint to a browser over SSE and handle disconnects and backpressure.
2	Retrieval and RAG Retrieval & RAG	You build hybrid retrieval (BM25 plus vectors with reranking) over a real corpus and measure recall before touching the generator.
3	LLM evaluation LLM Evaluation	You run an eval suite against two models on your own endpoint and report the delta with confidence intervals.
4	Model routing and cascades Model Routing & Cascades	You can design a cascade that sends easy traffic to a small model and show the cost and quality trade-off on a held-out set.