Production Claude Application Capstone
View badge details
About This Course
Ship a production Claude application. Environment config, observability, safety layers, deployment patterns (canary, feature flags), incident response runbooks, and the final end-to-end Orion assistant capstone that brings together every technique from CLD-AI-101 through CLD-AI-111.By the end of this course you will have built, deployed, and instrumented a production-grade Claude application for Orion Analytics one you can point at as your portfolio piece and reference architecture.
Course Curriculum
10 Lessons
Production readiness checklist
The 12-item checklist every Claude app should pass before shipping: secrets, evals, cost caps, logging, retries, safety, feature flags, alerts.
Build the app shell with config and healthz - Lab Exercises
Wire the Orion assistant baseline: typed config from env, a healthz endpoint, and a first Claude call gated on config validity.
Observability — structured logs, traces, metrics
Log format standards, trace_id propagation, per-request metric emission (latency, cost, cache-hit, refusal). What you can't measure you can't ship.
Instrument Claude calls with structured logs - Lab Exercises
Wrap the Anthropic client so every call emits one JSON log record with trace_id, latency, tokens, cost, cache hit, and refusal signal.
Safety layer — injection defense, PII, content filters
Prompt-injection detection on user input, PII redaction before logging, refusal patterns for out-of-scope requests, and post-hoc safety check on output.
Add the safety layer to the Orion app - Lab Exercises
Layer ingress injection detection + PII redaction on logs + output PII scrub. Verify 3 test cases behave correctly.
Deploy safely — feature flags, canary, rollback
Feature-flag model versions + prompts. Canary at 5% → 25% → 100%. Instant rollback via flag flip. Compare canary vs baseline metrics before promoting.
Ship a canary with metric-gated promotion - Lab Exercises
Route 5% of simulated traffic to a canary prompt variant. Emit per-variant metrics. Decide promote/rollback based on refusal + p95 delta.
Incident response — dashboards + runbooks
On-call playbooks for the 5 most common Claude-app incidents: cost spike, refusal spike, 429 storm, prompt-injection outbreak, upstream outage.
Orion end-to-end assistant capstone - Lab Exercises
The final capstone. Combine caching, safety, routing, structured logs, feature flags, PII scrub, and Haiku classifier into one production-shaped Orion assistant.
This course includes:
- 24/7 AI Instructor Support
- Live Lab Environments
- 5 Hands-on Lessons
- Completion Badge
Earn Your Badge
Complete all lessons to unlock the Production Claude Application achievement badge.
Production Claude Application
Awarded on completion of CLD-AI-112. The holder can ship a production Claude application with config, observability, layered safety, feature-flagged deploys, canary rollouts, and incident runbooks.
Skills You'll Earn
Complete all lessons in this course to earn this badge