Pulse: Production-grade unstructured document extraction

Pulse is a document extraction system that separates layout analysis from language modeling to address the limitations of generic OCR and VLM-based pipelines in handling complex documents.

Details

Pulse is a document extraction system developed by Sid and Ritvik, the co-founders of the YC S24 startup. They realized that while modern vision language models are good at producing plausible text, they can make mistakes in real-world document processing at scale. The core challenge is not extraction itself, but confidence - vision language models embed document images into high-dimensional representations optimized for semantic understanding rather than precise transcription, which can lead to subtle but meaningful errors, especially in tables and numeric fields. Pulse addresses this by separating layout analysis from language modeling, normalizing documents into structured representations that preserve hierarchy and tables, and tying extracted values back to source locations to surface ambiguity. This hybrid approach combines traditional computer vision techniques, layout models, and vision language models to handle complex document types more reliably.