Trajectory raises $45M to let domain experts keep training their own AI models

Today's models don't learn from their mistakes without more training. Companies adapting them for fields like law and finance usually fine-tune general models, which means hiring research teams or paying forward-deployed engineers[1]. Co-founder Ronak Malde compared a foundation model to "a very smart PhD student who's always starting a new job," adding that "what the economy values is someone who worked on the job for 30 years"[1].

Customers start with an open-source model and connect Trajectory to their application. The platform collects signals such as traces, corrections, edits and re-prompts, and Trajectory says it works with any agent harness and takes two code changes to set up[2]. Domain experts bring examples of good work, run training, compare the result against frontier models and push updates live[2]. Every update runs through the customer's own eval suite and approval workflow before it touches production, and each training run is auditable[2].

Harvey and Rogo already use the platform to post-train open models so they behave more like experienced lawyers or financial analysts. Other customers include Decagon, Mercor and Nvidia[1]. Trajectory's site quotes Nvidia CEO Jensen Huang at GTC pointing to Trajectory and Harvey post-training Nemotron[2].

Trajectory has 18 people and was founded about a year ago by three Stanford classmates. Malde was one of the researchers Google DeepMind acqui-hired from Windsurf, Michael Elabd also worked at DeepMind, and Arjun Karanam worked on generative foundation models at Apple[1]. Elabd argues that keeping humans in the training loop makes the approach safer than fully autonomous self-improvement[1].

It is entering a crowded field. TechCrunch names Thinking Machines Lab and its Tinker fine-tuning platform, Applied Compute, valued at $3.25 billion last month, and Prime Intellect as rivals automating post-training[1].

If specialists can tune open models on their own work without a research team, the edge shifts from whoever has the biggest general model to whoever owns the best feedback from real use. Trajectory is betting that legal, finance and support teams will want to own that loop themselves.