Trajectory raises $45M for no-code AI model training
Original: Trajectory raises $45M to make AI model training accessible to nontechnical experts
Why This Matters
Automating AI fine-tuning could cut the cost and time of deploying specialized models.
Trajectory, a Stanford-founded startup, raised $45M in a Series A at a $300M valuation led by Sequoia, with Nvidia, Bessemer, and Bain Capital joining, to automate AI fine-tuning for non-technical domain experts.
Trajectory, founded about a year ago by three Stanford classmates with backgrounds at Google DeepMind and Apple, wants to remove the expensive human layer that currently sits between a general AI model and a specialized one. Right now, businesses adapting models for law, finance, or medicine typically need internal research teams or costly forward-deployed engineers to do that fine-tuning work. Trajectory automates the loop: connect its platform to your application, collect feedback on model outputs, and let the system retrain the model continuously from that signal.
Co-founder Ronak Malde describes base models as 'a very smart PhD student who's always starting a new job,' while the real-world value lies in someone with 30 years on the job. That gap is what Trajectory targets.
The 18-person company's platform is now publicly available. Early customers include Harvey, Rogo, Decagon, and Mercor, each using it to post-train open source models on domain-specific tasks—reviewing legal contracts, valuing companies by a particular methodology. Nvidia is both an investor and a customer.
On safety, co-founder Michael Elabd draws a line between Trajectory's human-in-the-loop approach and recursive self-improvement: keeping specialists in the feedback chain, he argues, guards against unintended model drift. Trajectory's closest competitor is Thinking Machines Lab, the company backed by former OpenAI CTO Mira Murati.