Applied engineering
experience and perspectives
AI systems, delivery decisions, product operations and the mistakes worth writing down. Written by the senior team between engagements, not by a content calendar.
When the Automation Itself Becomes the Incident
Automations fail silently more often than loudly. The observability and recovery patterns that catch a broken workflow before a customer reports it.
Automating the Internal Workflow Without Creating a Second Job
Most internal automation moves the work from doing the task to maintaining the automation. Patterns that keep maintenance below the original task cost.
Temporal vs n8n vs Zapier: Honest Comparison
Three workflow tools for three different teams. Reliability, ergonomics, self-hosting, AI-in-the-loop. The decision matrix without tool-fashion bias.
Warehouse Cost Control: The Quarterly Ritual
Snowflake, BigQuery and Redshift bills grow quietly until the CFO asks why. The quarterly ritual that catches offenders before emergency cost-cuts.
Streaming Data Pipelines: The Back-Pressure Trap
Streaming pipelines work until the consumer falls behind the producer. Patterns that handle back-pressure without dropping data or stalling the source.
Feature Store Architecture: The Three Shapes
A feature store is a library, not a database. The three shapes that work in production and the engineering trade-offs that decide which one fits your team.
Model Monitoring Beyond Accuracy Drift
Accuracy drift is the easy signal to monitor and the late one. The five signals that catch a degrading model before customers notice the regression.
Shipping Classical ML in an LLM World
LLMs did not retire classical ML. The categories where gradient boosting and logistic regression still beat LLMs on cost, latency and reliability.