A personal project exploring why AI-generated SQL isn't always trustworthy and how semantic context, validation, and governance improve analytics accuracy.
AI-generated SQL can look right while being wrong. Learn how human-in-the-loop workflows build trust through reviews, approvals, audits, and escalation paths.
Run an offline Playground eval with a cross-family LLM judge and use failing rows to separate retrieval issues from generation problems and judge noise.
We optimized for code-generation speed while the real bottleneck — cognitive overhead and knowing where to make changes — remained completely untouched.
Automate Power Platform deployments by resolving connection references and activating cloud flows after import, so every environment is ready when the pipeline finishes.
LLMs can quickly generate web application code, but AI-written code may contain security vulnerabilities. This article reviews testing methods for LLM systems.
This post traces that journey using triangular number computation as a practical example of intentional fall-through and connects the technique to Duff's Device.
Learn Temporal workflow design patterns for reliable distributed systems using durable execution, sagas, polling, fan-out/fan-in, signals, and versioning.
Learn how to generate documentation using an LLM with mdship, and how to ensure that the prompts, which are now the source documentation, do not get lost.