Trajectory, a San Francisco startup that helps companies keep improving AI models after they are deployed, has raised $45m in a Series A round led by Sequoia Capital. The company announced the round on 2 November 2026 and said it valued the business at $300m.
Nvidia, Bessemer Venture Partners and Bain Capital Ventures also invested. The company used the announcement to make its post-training platform publicly available, with a self-serve version aimed at people who are not machine-learning engineers.
In August, Crypto Briefing reported a Sequoia-led round of $40m at a $300m valuation, citing The Information. The amount Trajectory announced in November was $45m.
“Imagine if everyone had the tools to shape intelligence itself.”Ronak Malde, Co-founder and CEO, Trajectory
What Trajectory does
Most companies that build AI products rely on models released by large labs and wait for new versions to get better results. Trajectory's platform takes a different route. A customer starts with an open-source model, connects it to its own application and collects feedback on how well the model performs. Trajectory then uses that feedback to train the model again, an approach often called continual learning.
Users describe the tasks they care about and how results should be scored, then train, test and deploy new versions of the model. The platform includes two features the company calls research previews. Continual Grading learns a customer's standards from a small number of examples so that fewer outputs need to be labelled by hand. Group Size 1 trains on real interactions after a model is in production. The company says customers control which data is used and that every training run can be audited.
Co-founder and chief executive Ronak Malde told TechCrunch that the people using the product decide what good work looks like for their own organisation, and that experts with decades of experience in a job are what the economy values.
Customers and early results
TechCrunch reported that Harvey and Rogo use Trajectory to post-train open-source models, and that Decagon, Mercor and Nvidia are also customers. Trajectory's website also shows Clay and Mistral AI.
In its launch post, Trajectory said one customer trained more than eight models for long, multi-step agent tasks in two weeks. It also said it post-trained Nvidia's Nemotron 3 Ultra model on legal tasks within 24 hours of the model's release. According to the company, a smaller Nemotron model it trained went from a 0% to an 8.3% all-pass rate on Harvey's LAB benchmark.
The team and its funding
Trajectory was started about a year ago by three Stanford classmates, according to TechCrunch. Malde was among the researchers Google DeepMind hired from Windsurf. Michael Elabd also worked at Google DeepMind, and Arjun Karanam worked on generative foundation models at Apple. The company has 18 employees, TechCrunch reported.
In May 2026 Trajectory raised $15m from investors including Conviction, Bessemer Venture Partners and Radical Ventures, with AI researchers Jeff Dean and Fei-Fei Li also backing the company, Pulse 2.0 reported. Trajectory's website now gives its total funding as $60m across its seed and Series A rounds.
The company did not say how it plans to spend the new money. It faces competition from other startups working on continual learning and custom training, including Thinking Machines Lab and Applied Compute, according to RuntimeWire.
Company website: trajectory.ai