The latest articles on DEV Community by Dave (@dave8172). https://dev.to/dave8172 https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F3987397%2F0f6b44d9-ee2f-48bc-95b1-d9433ded1ffd.png https://dev.to/dave8172 en Dave Tue, 16 Jun 2026 12:29:37 +0000 https://dev.to/dave8172/show-hn-doceval-eval-harness-for-llm-document-extraction-pipelines-3gd7 https://dev.to/dave8172/show-hn-doceval-eval-harness-for-llm-document-extraction-pipelines-3gd7 <p>I kept seeing the same gap: people ship LLM-based document extractors (invoices, receipts, forms) with no systematic way to know how accurate they actually are. So I built doceval — point it at your extractor function + a labeled dataset and get back field-level accuracy, a failure taxonomy (missed_field / hallucination / wrong_format / wrong_value), and optional per-document cost tracking.</p> <p>Works with any extractor (Claude, GPT, regex, rules) and any document schema. One JSON label file per document, one Python function, one CLI command.</p> <p>Includes a working 20-document invoice example with a Claude Haiku extractor so you can run it immediately.</p> <p>GitHub: <a href="https://github.com/dave8172/doceval" rel="noopener noreferrer"></a></p> ai development python opensource