EvalLens – Open-source tool to evaluate structured LLM outputs
Schema conformance checks beat generic text evals for JSON-heavy LLM pipelines.
RAIF — Repairable AI Interchange Format: a wire format for LLM-emitted JSON that round-trips losslessly, self-repairs syntax errors, and costs fewer tokens than JSON.
Self-repairing LLM output format that beats JSON on tokens and recovery.
Developers building LLM applications with structured outputs
Instructor · Outlines · jsonrepair
Schema conformance checks beat generic text evals for JSON-heavy LLM pipelines.
Finally separates JSON validity from actual value hallucination in LLM outputs.
Append-only lineage prevents LLM outputs from collapsing structure—but unclear if it ships or works.
Yet another code-to-LLM formatter when repomix and git2llm already exist.
Tree-sitter interface extraction cuts token usage by 6x, but chat context window optimization is becoming table stakes.
LiteLLM already does this with more providers, more features, and way more maturity.