Blog

Notes from shipping production AI — tooling, agents, evaluation, and lessons from real systems.

• When AI Gets Your Name Wrong — Speech models are fluent, but company names still defeat them — and fixing that means knowing when not to guess.
• The Secret Life of Words — How machines learned to speak our language: tokens, embeddings, attention and next-token prediction — and why fluency is not the same as truth.
• Jev: a model that decides instead of writes — TypeSafe’s System One model returns labels and probabilities, not essays — where it fits beside an LLM, and what independent numbers actually show.
• MCP when the model can call your APIs — A short tutorial, with Kong as the secure path: MCP is how a model proposes tool calls — the gateway decides what runs.
• Optuna when every trial is expensive — A short tutorial with a worked example: multi-objective TPE when each trial is a full train, and why holdout — not Optuna — decides what ships.
• Search Party: How Two Rival Ideas About Retrieval Learned to Share — For decades, search split into two camps: the literalists who matched words, and the semanticists who matched meaning. The peace treaty between them is now the quiet backbone of production AI.
• Graphify: Giving Your AI Coding Assistant a Map — Graphify builds a knowledge graph from your codebase so AI assistants stop re-reading every file. Here's how it works and where it breaks.
• Deploying a Vite SPA on GoDaddy PaaS — What we ran into deploying a Vite React app on GoDaddy's Node.js hosting, and how we fixed it.
• Multimodal Video Search: From Notebook to Production — How we built a CLIP-based scene search system that cut video lookup time from minutes to seconds.
• Five Things We Learned Shipping RAG Systems — What actually mattered when we shipped RAG to production, and what didn't.