Ocular AI

Ocular AI builds a multimodal data lakehouse, called Ocular Foundry, for turning raw video, image, audio, and text data into labeled datasets used to train and evaluate AI models. The platform lets teams ingest large volumes of unstructured data, search across it, annotate it with human-in-the-loop review, version datasets over time, and then train or evaluate models using built-in GPU access — all within one collaborative workspace rather than stitching together separate tools. A companion service, Ocular Bolt, offers on-demand, expert-led labeling that combines AI-assisted automation with human subject-matter experts and reinforcement-learning-style feedback for cases that need higher accuracy. The company also offers datasets built specifically for speech and conversational AI, including full-duplex, two-speaker recordings with natural disfluencies preserved, domain-specific sessions such as medical intake or customer support calls, and scripted single-speaker voice recordings, spanning more than 40 languages. Ocular AI is built for teams developing conversational AI, voice agents, speech recognition, and text-to-speech systems who need higher-quality training data than generic datasets provide. The company was founded by engineers from Microsoft and Google, is based in San Francisco, and is backed by Y Combinator and Drive Capital. Pricing is not published; interested teams can request samples directly.