Compiler-driven feedback loops force LLMs into 100% schema compliance on complex types.
Backend engineers building AI agents or code generation pipelines
Instructor · Guidance · Pydantic
The talk was about how I got function calling to work reliably on deeply recursive union types — the stuff the industry generally says doesn't work. With qwen3-coder-next, first-try success rate was 6.75%. And the entire Qwen 3.5 model family was hitting 0% on union types due to a consistent double-stringify bug. Both ended up at 100%.
Slides are also downloadable in the article — speaker notes are written inside as slide notes if you'd like the full narrative behind each slide.
TypeScript only alternative to Instructor: compiler validation forces 100% function calling success.
TypeScript compiler for LLM validation beats Instructor and Guidance on nested schemas.
Compiler-level validation turns Qwen's 6.75% structured output success rate into 100%.
Self-bootstrapping agent writes its own improvements in 100 lines of TypeScript.
Deterministic policy matrices block AI agents from executing dangerous API calls.
TypeScript in Postgres beats plv8's plain JavaScript with actual type safety.
