ai-explained

case studies · whole products · what you'd ship

Real products from the /ship stack.

/build taught you the model. /ship taught you the production stack. These case studies show what real products look like when you compose that stack — architecture, the code unique to each product, the eval story, and an honest retrospective. Every study cross-references the /ship steps it builds on, so you can plug back in to the foundations any time.

5
case studies
5
ready to read
2h 24m
of reading
6h 45m
of hands-on
who this is for

If /ship taught you the parts, this teaches the products

if you've finished /ship
This is the application layer
/ship gave you the production stack — server, RAG, tools, agent loop, observability. Case studies show what real products look like when those pieces compose, with the trade-offs visible.
if you're planning a build
Steal the architecture
Each study is a self-contained walkthrough — diagram, code, evals, retrospective. Closer to a "what would I do, given the budget" than a tutorial.
if /ship is too theoretical
Ground it in real shapes
Case studies are concrete: a real product, real eval numbers, real what-broke notes. They make /ship's abstractions click in a way that bullet points can't.
the studies

5 products, four shapes

Each study targets a different product shape: retrieval-heavy, agent-heavy, multi-agent-heavy, all-of-the-above. They're orderable but not ordered — pick whichever shape your product looks most like.

  1. 01
    Docs assistant with citations
    RAG over a real corpus, strict citation requirements, zero hallucination tolerance. The product /ship/06–08 wants to be.
    28 min + 1h hands-on /ship/05, 06, 07, 08, 12, 13
  2. 02
    Code-review agent
    Reads a PR diff, runs tests, comments inline, produces a verdict. The product /ship/09–10 wants to be.
    26 min + 1h 15m hands-on /ship/09, 10, 12, 13
  3. 03
    Research assistant
    Multi-agent fan-out for cited briefs. The product /ship/11 wants to be — with the cost/latency trade-offs in real numbers.
    28 min + 1h hands-on /ship/08, 09, 10, 11, 12, 14
  4. 04
    Customer-support bot
    RAG + a tools layer that hits a (mock) CRM + escalation logic. The product that composes the entire /ship stack.
    32 min + 1h 30m hands-on /ship/05, 06, 07, 08, 09, 10, 12, 13, 15
  5. 05
    The cheapest version of itself
    Take the docs assistant from CS-01 and distill it into a 7B student. Same retrieval, same citation contract, ~7× cheaper, ~5pp eval gap. The product that pays for itself.
    30 min + 2h hands-on /ship/04, 13, 14, 17

How this fits the rest of the site

Case studies are the third leg. /build teaches the model from scratch; /ship teaches the production stack; /case-studies shows what real products look like when the stack composes. Where /ship is "step by step from nothing to a deployed service," each case study is "here's a finished product, here's what made it hard, here's the code unique to it."

Articles cross-link aggressively into /ship for foundations and /articles for theory, so you don't have to re-read everything to follow along. If something seems too compressed, the link to the underlying step is one click away.

Status: 5 live · 0 wip · 0 stubbed.