IDP AutoOpt slashes document pipeline tuning from weeks to under 2 hours

A new research paper presents IDP AutoOpt, an LLM-based agent that autonomously discovers optimal configurations for intelligent document processing (IDP) pipelines. Currently, tuning prompts, models, OCR settings, and schemas jointly requires domain specialists 20 to 80+ person-hours per document type—a bottleneck that grows as enterprises add more document classes. IDP AutoOpt runs a closed loop: it scores a configuration on a small labeled set, diagnoses field-level errors, generates targeted edits, and re-evaluates, guided by human-authored domain skills that encode production expertise.

The results are striking. On an extraction benchmark, the agent achieves 90.2% accuracy compared to 81.6% for human experts, while reducing per-page cost by 4.6x. Configuration time collapses from weeks to under two hours. The researchers also share practical lessons: LLM capability has a hard threshold below which optimization fails, and curated domain skills outperform raw source-code access, which can degrade performance without structure. The approach is not limited to IDP—it extends to other enterprise AI systems like RAG and multi-agent workflows, promising to unblock a key deployment bottleneck.