Pulse

Pulse builds production-grade document intelligence infrastructure, using OCR, layout detection and vision-language models to turn complex unstructured documents into structured, LLM-ready data at scale. Its platform runs a multi-stage extraction pipeline: specialised component detection models, low-latency OCR, reading-order algorithms and table structure recognition, with fine-tuned vision-language models handling charts and figures. Target document types include financial statements, medical records, legal contracts and real estate documents.

Technical work spans several established problem areas at once: table structure recognition, reading-order algorithms, component detection, and the structuring of extracted text for downstream large language models. The stated emphasis is on enterprise-grade accuracy and reliability, backed by deployment options in private VPCs and on-premises environments for clients with stringent security requirements.

The Pulse Document Intelligence Platform has processed over 1 billion pages for clients. Its customer base ranges from Fortune 100 enterprises to Y Combinator startups, and it serves verticals including financial services, healthcare, legal, real estate and enterprise software.