2 ms·

Show HN: AI-Powered Structured Data Extraction from Any Document (93%+ Accuracy)

Hey HN! we built Structurify (https://structurify.ai https://structurify.ai) to solve a problem we kept hitting at our AI consultancy: enterprises drowning in unstructured documents that need to become structured data. The typical approach is OCR + custom ML pipelines, which takes months to build and degrades on anything that isn't a clean PDF. We took a different approach — contextual AI understanding instead of character-level OCR. What it does: Upload any document (PDF, scan, photo, Word, Excel, PowerPoint) → describe what you want extracted in plain English → get structured JSON/CSV in ~30 seconds. 93%+ accuracy. We have 169 pre-built extraction templates (SDS, invoices, medical records, W-9s, construction pay apps, etc.) but you can also describe custom extractions without any template. Some things we're proud of: Multi-page understanding (connects data across 50+ page documents) Multi-language (30+ languages) Confidence scoring with human-in-the-loop review Full REST API, webhook support, most integrations done in a day $0.20/extraction, no subscription, credits never expire Curious what HN thinks. Happy to answer technical questions about the architecture. Free trial: 50 credits with work email, no CC required.

ares623 9mo ago
What does 93% accuracy mean here? My understanding is that tech like LLMs are not able to give a confidence score unlike traditional OCR methods. How was 93% calculated? How are the parsing errors surfaced?
- chaitanyavelaga 9mo ago
  Great question. The 93% refers to field-level accuracy on a labeled test set; a field is counted as correct only if it matches the ground truth (with basic normalization for dates/currency, etc). It’s not document-level accuracy. We don’t use raw LLM probabilities for confidence. The score is based on additional validation checks (cross-field consistency, format rules, reconciliation). Each field is returned with its confidence and any validation flags so errors are visible and reviewable.

ares623 9mo ago

What does 93% accuracy mean here? My understanding is that tech like LLMs are not able to give a confidence score unlike traditional OCR methods. How was 93% calculated? How are the parsing errors surfaced?

chaitanyavelaga 9mo ago
Great question. The 93% refers to field-level accuracy on a labeled test set; a field is counted as correct only if it matches the ground truth (with basic normalization for dates/currency, etc). It’s not document-level accuracy. We don’t use raw LLM probabilities for confidence. The score is based on additional validation checks (cross-field consistency, format rules, reconciliation). Each field is returned with its confidence and any validation flags so errors are visible and reviewable.