List of lectures for the course ChatGPT Apps, 17 quest level on JavaRush

Structured logs and request correlation
ChatGPT Apps
17 level,0 lecture
In this lecture, we break down how to turn scattered messages into an observable system: we introduce structured JSON logs, a single trace_id for end-to-end correlation (ChatGPT App → MCP → widget → backend → webhooks), a minimal “standard” set of fields ( service, env, flow, step), logging practices for tools and the backend proxy, and security rules (PII scrub, secrets, limiting DEBUG in production).
Open
Metrics and SLO: p95, error‑rate, webhooks, availability
ChatGPT Apps
17 level,1 lecture
In this lecture we cover basic observability for a ChatGPT App using GiftGenius as an example: why “seems to work” must be replaced with measurable SLI and targets SLO, how to read and use percentiles ( p50/ p95/ p99) instead of the average, how to calculate error-rate and availability, why webhooks are a separate risk area for commerce, and how to capture metrics in TypeScript/Next.js via a simple logMetric.
Open
Testing GiftGenius — unit, contract, E2E, and smoke in CI
ChatGPT Apps
17 level,2 lecture
In this lecture we assemble a “minimum viable” testing pyramid for a ChatGPT App using GiftGenius: fast unit tests for pure functions ( scoreGift, rankGifts), contract tests with Zod/JSON Schema for MCP tools ( inputSchema/ outputSchema) and ACP/webhooks, “almost E2E” without chatgpt.com (React Testing Library + Playwright), and smoke checks in CI to guarantee: Next.js starts, MCP responds to basic calls, and checkout endpoints don’t crash on a typical JSON.
Open
Release process — versions, notes, SDK/spec migrations, feature flags, rollback
ChatGPT Apps
17 level,3 lecture
In this lecture we assemble a practical release process for the ChatGPT App: introduce a version matrix by layers (code, tool schemas, prompts, data), document CHANGELOG and release notes, learn to safely migrate SDK/ spec, separate deploy from release via feature flags, and build a clear rollback plan — including “dangerous” changes to the manifest and tool schemas.
Open
Lightweight load tests and feed data quality
ChatGPT Apps
17 level,4 lecture
In this lecture, we cover how to run “lightweight” load runs for a ChatGPT App and MCP/ACP backend: why concurrency matters more than abstract RPS, how to control p95/ p99 latency and error rate, and why you should test your endpoints directly ( /api/mcp, /api/checkout/create) rather than through the ChatGPT UI. Then we switch to product feed quality: structural validation via Zod, business rules, and LLM‑friendly data normalization. Finally — how to bake smoke‑load and feed checks into the release process and which mistakes teams most often make.
Open