Insight · Plateer AI Labs

Article series

The core ideas you need to make sense of an LLM, taken one at a time.

Three weeks of reading 80 defect reports and rebuilding the code, the errors, and the manual

Designing the execution environment an AI agent works in, from the ground up

Starting from the questions, building a knowledge graph, then working on quality and search

As builds stretched into hours we replicated job state into a shared cache. When it expired we could not recover cancellation and resume state.

We packaged a workflow as a wheel and installed it elsewhere. Calling a tool exposed every assumption that only held inside the product.

Three weeks of GS certification testing produced 80 defects, but fewer than half were code bugs. The rest were behaviours a user could not confirm.

We added multi-turn search to follow relations. Context grew, so we cut tool results to 500 characters, and long sources never reached synthesis.

The engine had its own module, and running it still needed the product's config object, DB session, and canvas classes. Closing a real boundary.