floatingpoint

Floatingpoint builds high-fidelity environments, evals, and supervised datasets to advance visual intelligence. They provide data products and services to create and evaluate frontier vision capabilities, with a focus on long-horizon tasks and domain-specific datasets.

The company creates high-quality supervised visual datasets and evaluation environments across domains such as document layout, form controls, table understanding, schematic analysis, 3D geometry, and long-horizon tasks. Their platform supports end-to-end data production (data signals, QA, annotation tooling, ML-assisted systems), delivery pipelines with QA clearance, and ML-backed evaluation. They offer off-the-shelf data packets, benchmarks, and bespoke services including co-design of schemas, annotation guides, golden datasets, and production-scale data distribution, all integrated through their platform and marketplace.

Who it’s for: Startups, enterprises, and model labs working on visual intelligence and document/vision tasks requiring high-quality supervised data, benchmarks, and custom datasets.

Hiring and research-focused posts; YC backing; partnerships with startups, enterprises, and model labs