Trajectory raises $45M to make AI model training accessible to nontechnical experts

Trajectory, a startup founded about a year ago by three Stanford classmates with experience at Google DeepMind and Apple, has raised $45 million in a Series A round at a $300 million valuation. Sequoia led the investment, with participation from Nvidia, Bessemer Venture Partners, and Bain Capital Ventures. The company, which has 18 employees, is now making its post-training platform publicly available.

The pitch rests on a limitation shared by models from OpenAI, Anthropic, and open-source projects alike: they don't learn from their mistakes without further training. Businesses adapting models to fields like law and finance usually handle this by fine-tuning, which often means hiring internal research teams or paying consultants known as forward-deployed engineers. Co-founder Ronak Malde compared foundational models to a very smart PhD student who is always starting a new job, adding that what the economy values is someone who worked on the job for 30 years.

Trajectory's answer is an approach called continual learning. Customers start with an open-source model, connect the platform to their applications, and let it collect feedback on model performance; Trajectory then uses that data to retrain the model so it improves at specialized tasks over time. Malde said lawyers, doctors, and other specialists can improve the models near automatically themselves, which could make deploying specialized AI faster and cheaper. He also said customers decide what good looks like for their organization and turn that into model improvements.

Existing users include Harvey and Rogo, which post-train open-source models to behave more like experienced lawyers or financial analysts by learning from prior tasks such as reviewing a particular type of legal contract or valuing a company under a specific methodology. Other customers listed are Decagon, Mercor, and Nvidia.

Co-founder Michael Elabd drew a line between Trajectory's method and recursive self-improvement, the idea that AI systems could improve their own capabilities. He argued that keeping humans involved in the training process makes Trajectory's approach safer, since autonomous self-improvement could produce unintended or dangerous outcomes.

What to watch now that the platform is public: whether nontechnical specialists can sustain the feedback loop without research staff behind them, and whether continually retrained open-source models hold up on the reliability and safety measures enterprises expect. Also open is how Sequoia and Nvidia's involvement shapes distribution, and how the human-in-the-loop safety argument holds up as the company pushes toward more automated retraining.