Trajectory, an 18-person startup founded about a year ago by three Stanford classmates, said Monday it has raised $45 million in a Series A round at a $300 million valuation, led by Sequoia with participation from Nvidia, Bessemer Venture Partners and Bain Capital Ventures.
The company is selling a way to keep post-training in the hands of the people who know the work. Its platform connects to a customer's application, watches how a base — usually open-source — model performs, collects feedback on what a good answer looks like, and periodically retrains the model on it. The pitch is that a law firm or a hospital can improve a model in its own domain without hiring a research team or renting forward-deployed engineers from a vendor.
Co-founder Ronak Malde's framing: foundation models are "a very smart PhD student who's always starting a new job," while "what the economy values is someone who worked on the job for 30 years."
Harvey and Rogo are already using the platform to post-train open-source models for legal and financial work; Decagon, Mercor and Nvidia are also named as customers. Trajectory is making the platform publicly available now, having previously run it with a set of design partners.
The technical bet is on continual learning — models that accumulate corrections over time rather than being frozen after a single fine-tuning run. That puts Trajectory in a crowded and suddenly expensive neighbourhood. It names Thinking Machines Lab, Mira Murati's company behind the Tinker fine-tuning platform, as a direct competitor, alongside Applied Compute — valued at $3.25 billion last month — and Prime Intellect.
Co-founder Michael Elabd draws a line between this and recursive self-improvement, the idea of models improving themselves without human involvement. Keeping domain experts in the loop, he argues, is what makes continual learning safe as well as useful: the human decides what "better" means before the model optimises for it.
The founding team's pedigree is part of the sales pitch. Malde was among the researchers Google DeepMind acqui-hired from Windsurf; Elabd also worked at DeepMind; Arjun Karanam worked on generative foundation models at Apple. The company has also hired from OpenAI, Stripe and Netflix, and puts unusual weight on design for a training company — the argument being that a post-training interface only democratises model improvement if non-specialists can actually use it.
Whether that is a durable moat is the open question the $300 million valuation is answering in advance.
