Trajectory announced a $45 million Series A round at a $300 million valuation, aiming to make post-training and continual learning accessible to nontechnical domain experts. The 18-person startup, founded about a year ago by three Stanford classmates with experience at Google DeepMind and Apple, says it will let organizations refine open-source models using real-world feedback from their own applications.
How the platform works
Customers connect an initial open-source foundation model to Trajectory’s platform so it can collect feedback on task performance. Trajectory then uses that feedback to retrain the model, a process the company describes as continual learning. Co-founder Ronak Malde compared foundation models to “a very smart PhD student who’s always starting a new job,” adding that the economy often values someone who has worked in a role for decades.
Trajectory says this post-training approach enables specialists — such as lawyers or financial analysts — to improve a model’s behavior for specific tasks without hiring internal research teams or expensive forward-deployed engineers. Early customers include Harvey, Rogo, Decagon, Mercor and Nvidia, with use cases like contract review and industry-specific valuation methodologies.
Team, safety and competition
The startup’s co-founders include Ronak Malde, Michael Elabd and Arjun Karanam. Malde was among researchers acqui-hired from Windsurf, Elabd previously worked at Google DeepMind, and Karanam worked on generative foundation models at Apple. Trajectory has also hired researchers and product designers from OpenAI, Stripe and Netflix. Karanam said the company developed an interface intended to make post-training more accessible to non-technical users.
Trajectory’s approach acknowledges similarities with concepts such as recursive self-improvement, but co-founder Michael Elabd argues putting humans in the loop reduces safety risks associated with autonomous self-improvement. The company identifies several competitors in the post-training and fine-tuning space, including Thinking Machines Lab (founded by Mira Murati), the team behind the Tinker fine-tuning platform, Applied Compute (valued at $3.25 billion last month) and Prime Intellect.
The Series A was led by Sequoia with participation from Nvidia, Bessemer Venture Partners and Bain Capital Ventures. Trajectory says its post-training platform is now publicly available and that customers determine “what good looks like” for their organizations as they apply the tool to specialized tasks.
Original source: TechCrunch AI