Research & Insights

Production-grade document intelligence accessible to everyone. All six primitives in a visual playground - no code required, no setup, no infrastructure to manage.

What if you could edit PDFs like you edit code? We're building Cursor for PDFs - upload a form, paste your context, and let AI auto-fill intelligently.

Why specialized models beat general-purpose frontier models on privacy, cost, and domain accuracy. Deep dive into Sentinel-PII, Precis, and Cernis-Legal-OCR.

Why do OCR systems extract text perfectly yet fail to understand what they're reading? Agentic OCR processes documents the way humans do through layout analysis, dual OCR, and structured output.

How we built document AI infrastructure capable of handling 1 million documents per day through component isolation, stateless workers, and tiered GPU compute.

How we built production-grade OCR and reasoning models using consumer hardware, efficient fine-tuning, and reinforcement learning for under $50 in GPU time.

Six primitives that transform unstructured documents into production-ready data. Deep dive into ocr(), extract(), classify(), summarize(), chunk(), and count_tokens().

Lightweight, accurate PII detection model fine-tuned on IBM Granite 4.0 that identifies and tags 20 different categories of sensitive information.

Open-source vision language model trained with reinforcement learning to understand and reason about documents, handling math, LaTeX OCR, invoices, and handwriting.

Unified OCR solution handling mathematical formulas, handwritten notes, and structured invoices in a single model using Qwen2.5-VL architecture.

Specialized vision-language model fine-tuned for legal document processing, trained on synthetic documents from the Caselaw Access Project.

Fine-tuned document intelligence model built on IBM Granite that's fast, runs locally, and maintains document privacy for on-premise processing.