floatingpoint

Post-training data to teach models document work

Y CombinatorWinter 2024ActiveB2B

What they do

Problem

AI models often underperform on real-world document workflows due to lack of high-quality post-training data.

Solution

Floatingpoint creates off-the-shelf, human-crafted and synthetically expanded post-training datasets validated through internal training cycles to improve model performance on document tasks.

Who it's for

AI/ML teams and companies building or fine-tuning models for document-centric workflows.

Founders

Mehul Chadda

Co-founder & CEO

Co-founder & CEO at floatingpoint I have a background in metrology and a bsc in materials engineering. I work on improving vision capabilities in models.

LinkedIn · X