Deployment, environments and releases
CI/CD and smoke tests
CI/CD pipelines and smoke tests
- How to Set Up CI/CD for a Small Team
- Contract Tests Between Your Smoke Suite and Its Vendors
- Splitting a Smoke Suite by Blast Radius: Read-Only, Writes, Money
- Writing a Smoke Test Suite the Pipeline Runs on Every Deploy
- A Canary That Reads Its Own Smoke Checks
- A Vendor Change Log: Fields, Cadence and Escalation
- Keeping a Flaky Test Suite from Blocking Every Deploy
- Naming the Assertion That Fired: Result Files for CI
- All of CI/CD and smoke tests
Preview and staging
Preview and staging environments
Test data and fixtures
Test data and fixtures
- Seeding a Staging Database with Fixtures and Anonymized Data
- Seeding a Vector Store for RAG Tests Without Drift
- A Smoke Probe That Writes Then Deletes: Keys and Teardown
- Anonymizing Production Data for Model Evaluation
- Deciding What a Smoke Suite May Touch: A Data Ownership Map
- Ephemeral Test Data: Generating and Tearing Down Fixtures per Run
- Eval Fixtures That Call Real Sandbox APIs: Cost and Cleanup
- Generating Realistic Test Data That Keeps Foreign Keys
- All of Test data and fixtures
Edge functions and regions
Edge functions and regions
- Edge Functions Explained: What Runs Where
- Choosing an Edge Region: Latency, Data Residency and Cost
- Edge Function Cold Starts: Causes and How to Reduce Them
- Giving an Edge Function a Model Call: Timeouts and Fallbacks
- How to Move a Database Query to the Edge Safely
- Caching Model Responses at the Edge: Keys, TTLs and Invalidation
- Choosing a Serverless Runtime for AI Workloads
- Sizing a Queue Behind an Edge Function's Write Path
- All of Edge functions and regions
Inference routing
Inference routing and gateways
- A Routing Rule That Sends Overflow Traffic to a Cheaper Model
- Gateway, Sidecar or In-Process Routing for Model Calls
- Routing by Difficulty: Picking the Smallest Model That Works
- Routing Inference Requests by Cost and Latency
- A Policy Table in Postgres: Schema and Cache
- A Routing Rule That Fires on Latency Before It Fires on Cost
- Draining Connections and Warming Models on Deploy
- Multi-Region Failover for AI Features: What Degrades First
- All of Inference routing
Prompt and model releases
Prompt and model releases
- A Version Id That Survives a Prompt Hotfix
- Feature Flags for Model Swaps: Pinning and Canaries
- Linking a Prompt Version to Every Model Call
- Separating Prompt Changes from Code Deploys
- A Prompt Compressor That Keeps the Facts
- A Rollback Plan for a Prompt or Model Change
- A Variable Rename in a Prompt Template: Finding Broken Calls
- Canary Metrics for an Agent-Backed Feature
- All of Prompt and model releases