RadarMakers editorial
Trajectory raises $45M to let experts retrain AI on real work
Trajectory wants lawyers, doctors, and other specialists to shape how AI models improve at their jobs. Its $45 million Series A backs a platform that is already being tested by enterprise customers.
Most AI models do not learn from a mistake just because a user corrects them. Trajectory wants to turn those corrections into a training signal, and it has raised $45 million to build that approach into a business.
TechCrunch reported on November 2 that the 18 person startup raised a Series A at a $300 million valuation, led by Sequoia with Nvidia, Bessemer Venture Partners, and Bain Capital Ventures participating. Its founders describe the product as a way for companies to connect an open model to their applications, collect feedback on its work, and use that feedback to retrain it for specialized tasks. The report says Harvey and Rogo use the platform for legal and financial work, and names Decagon, Mercor, and Nvidia as other customers. Read the funding and customer reporting from TechCrunch.
The bet is on letting people who know a job define a good result, rather than making every improvement depend on a specialist machine learning team. That is an ambitious promise for fields where errors carry real costs. The evidence still needs to catch up with the pitch. Trajectory's homepage labels Continual Grading and Group Size 1 as research previews and displays "$60M Series A + Seed." It does not show independent performance results or pricing. This company wide funding figure is broader than the $45 million Series A amount in the TechCrunch report.
The interesting question is whether expert feedback can make models reliably better on narrow tasks without turning every customer into a model training operation. A large valuation is not proof that it works. Customer published results will matter more than the funding headline.