Lectures

This lecture is about how to turn LLM and infrastructure spending into manageable metrics: where exactly to capture usage data (MCP server, agent steps, commerce), how to log it in JSON with linkage to request_id/ trace_id, how to estimate cost via tokens_in/ tokens_out and the cost_estimate_usd field, and which aggregates and dashboards help you quickly find “expensive, slow, and no conversion”.

In this lecture we’ll break down monetizing an LLM app using GiftGenius as an example: what models exist (B2C/B2B, freemium, subscription, upsell), how to tie price to cost_per_task, and how to run “cost ↔ quality” A/B experiments with proper logging ( experiment_id, variant, cost_per_task_usd, conversion, revenue). We’ll also outline the role of a future quality_score and why optimizing only for cost almost always breaks business metrics.

This lecture is about growing a ChatGPT App through product events: we translate marketing into the funnel language AARRR, design an event model ( "app_opened", "workflow_completed", "checkout_success"), link acquisition/activation/retention/revenue with analytics and cost instrumentation, learn to run campaigns as metric-driven experiments, and do not break privacy-first constraints.

In this lecture you will build a practical “ops” for ChatGPT Apps using GiftGenius as the example: what to treat as an incident (including quality incidents, when the model doesn’t call suggest_gifts), how SLO/SLI turn into alerts (for example, on checkout_success_rate), what the incident lifecycle looks like (detection → triage → mitigation → resolution → post-mortem), which roles are needed even in a one-person team, why to write runbooks, how to enable a degraded mode via feature flags, and how to keep the service “in shape” through an operational calendar.

In this lecture we assemble an “App passport” as a practical 1–2‑page document for production: we capture architecture (ChatGPT → Widget → MCP → Agents → ACP/Stripe), security and privacy (OAuth, scopes, PII, retention, secret rotation), observability and SLO, economics and product metrics, as well as operational scenarios (incidents, runbooks, on‑call). Finally — scripts for a technical and a product demo and a list of common mistakes when preparing PASSPORT.md.