Perceptual Labs applies a single training capability across several separate data jobs. It trains models on a company's catalog and documents, including embedding models trained on a catalog, documents, and jargon for search, retrieval, and recommendations and agents trained with reinforcement learning in a company's own environments that plan ahead and carry multi-step work through to done. Its models also understand and generate across every modality a business runs on, from text and documents to images, audio, video, 3D, and sensor data.
Perceptual Labs reads, extracts, and classifies scans, forms, and invoices. A model trained for one job is cheaper to run and faster on that job than a general model, while frontier models are generalists that fall short on the specifics. It serves companies with a lot of proprietary data and no machine-learning team, naming insurers and logistics, with healthcare and legal as examples.