- Author
- Nanonets (maintainers of Nanonets-OCR and docstrange)
- Published
- Undated
- Kind
- Landing page of an online developer handbook
- Original
- nanonets.com/cookbooks/structured-llm-outputs
- Subject
- Harnesses and agent loops
What it is
Nanonets wrote the handbook for developers who use LLMs in programs. The page says it covers how structured generation works, which tools to pick and quality.
What it says
- The handbook says LLMs usually produce valid JSON, XML or code but can occasionally fail.
- It says that without structured outputs, agents and automation are a pipe dream.
- It describes itself as a living document that updates regularly.
What evidence it gives
The page is short and offers only claims and an outline. It gives no examples, code or measurements.
Interests and limits
Nanonets maintains OCR models and an open-source document library, and the page says so. Only the handbook's introduction page was read, not its chapters on constrained and unconstrained generation. The advice in those chapters is not described here.