Perit.AI

Perit is a training-data company for frontier AI labs. Everything we deliver is recorded, transcribed, labelled and graded by people who actually do the job: ex-contact-centre QA leads, claims handlers, ward clerks and shopkeepers, recruited for the sector and locale each program needs. What we deliver: - Voice and speech data: 4,000+ hours of audio delivered across 9 locales, transcribed verbatim and word-aligned, with 95%+ QA-sample accuracy. - RL environments and benchmarks: speech-to-speech RL environments plus coding and RL task sets, currently 100 RL tasks per week and scalable to 2,000. - Embodied data for robotics: egocentric video, depth, teleoperation and handheld-gripper demonstrations recorded in real homes, shops and workplaces. - Data collection and annotation: video, image and text collected to spec, and annotation of any data type on our own Foundry workspace. Every batch ships with a quality report, consent references and a reviewer chain. We work with almost all frontier labs and have quickly gotten past $9M in revenue. Founded by 2x YC founders Kshitij Dixit (Zeo, YC W20) and Poorva Dixit (ex-Google Cloud AI, Uber, Atlassian), with a deep IIT network of researchers at top labs.

YC batch

F26 · Fall 2026

Status, per YC

Active

Category

AI, Data

Team size

47

Location

San Francisco, CA, USA

Founded

2026

Checked

10h ago

YC tags: Data Labeling