Ship
Instrumented client
A small library: schema, retry, stream, fallback, logs.
Treat the model as a flaky network service: timeouts, retries, streaming, fallbacks, cost per request, and a dashboard.
Created by Baljeet Dogra
Ship
A small library: schema, retry, stream, fallback, logs.
Ops
Tokens, latency, £ per request — live, not a spreadsheet after the bill.
Expand a part for the syllabus. Content stays searchable when closed.
Auth, models, tokens, rate limits. What the SDK hides.
Timeouts, backoff, jitter, idempotency, streaming and cancellation.
JSON schemas, repair, a second provider, graceful degradation.
Attribution, budgets, a dashboard. Capstone: the client in a real endpoint.
You know HTTP. The model still looks like magic in the business logic.
You call one SDK method. Production will not forgive that.
Related: Prompt Engineering Mastery · Software Engineering Fundamentals
At least one frontier API and a second path so fallback is real. The course stays vendor-plural on purpose.
This is sprint 1 of that idea, in four weeks. The 28-week course continues through evals, RAG, agents and hardening.
Four weeks to an LLM client you could put on call. Create an account to enrol.
Enrol now