IDP AutoOpt Agent Automates Document Processing Pipeline Optimization.

Configuring intelligent document processing (IDP) pipelines is a labor-intensive and costly process, often requiring domain specialists weeks to tune prompts, models, OCR settings, and schemas for each document type. This bottleneck prevents enterprises from scaling their IDP solutions across numerous document classes. IDP AutoOpt introduces an autonomous LLM agent designed to automate this optimization. The agent operates in a closed loop: it scores a configuration against a small labeled dataset, identifies field-level errors, proposes targeted adjustments, and then re-evaluates the pipeline. This process is guided by human-authored "domain skills" that embed production expertise. In evaluations across healthcare, marketing intelligence, and financial services, IDP AutoOpt matched or surpassed human expert accuracy while drastically cutting configuration time from weeks to under two hours. For example, it achieved 90.2% accuracy versus 81.6% for human experts at 4.6 times lower per-page cost on an extraction benchmark. The research also highlights that agent LLM capability has a critical threshold for successful optimization and that structured domain skills are more effective than raw code access. The approach is generalizable beyond IDP to other enterprise AI systems facing configuration challenges.