Fine-tuning a model for a law firm or a hospital usually means hiring researchers or calling in forward-deployed engineers. Trajectory is selling the opposite arrangement.
The 18-person company raised $45M in a Series A this week at a $300M valuation. Sequoia led, with Nvidia, Bessemer Venture Partners and Bain Capital Ventures participating.
The claim under the pitch
The product wraps post-training in an interface built for domain experts. A customer connects the platform to its application, collects feedback on model output, and the system retrains an open model on that signal.
Ronak Malde, one of three Stanford classmates who founded it, compares a base model to a brilliant PhD student starting a new job every morning. What an economy pays for is experience in a role.
Harvey and Rogo use it to make open models behave more like lawyers and analysts. Decagon, Mercor and Nvidia are also customers, and the platform opened publicly this week.
The crowded middle
Trajectory is not alone in automating post-training. Thinking Machines Lab, Applied Compute and Prime Intellect are all chasing it, and Applied Compute was valued at $3.25B last month.
Co-founder Michael Elabd positions human involvement as the differentiator. Continual learning resembles recursive self-improvement closely enough that the distinction is a policy choice rather than a technical one: feedback comes from people who decide what good output looks like.
Malde came out of DeepMind’s acqui-hire of Windsurf, Elabd from DeepMind, and Arjun Karanam from generative models at Apple.
