Trajectory, a startup that wants AI models to learn from their mistakes the way people learn on the job, has raised $45 million in a Series A round at a $300 million valuation. The round was led by Sequoia, with participation from Nvidia, Bessemer Venture Partners and Bain Capital Ventures, TechCrunch reported on November 2.
Founded about a year ago by three Stanford classmates — Ronak Malde, Michael Elabd and Arjun Karanam, with experience at Google DeepMind and Apple — the 18-person company is betting on a gap that every business using AI runs into: a powerful general-purpose model is brilliant on day one and strangely forgetful by day thirty, because it does not learn from its own errors without further training.
The PhD student who keeps starting over
Co-founder Ronak Malde compared foundation models to a very smart PhD student who is always starting a new job. However capable the model is in general, it arrives at each company knowing nothing about that company's work — its contracts, its valuation methods, its house style — and it stays that way unless someone retrains it.
Trajectory's answer is an approach known as continual learning. A business starts with an open-source model and connects Trajectory's platform to its applications, which collect feedback on how the model performs in real work. That feedback is then used to retrain the model, so it gradually improves at specialized tasks instead of repeating the same mistakes.
Post-training for nontechnical experts
The pitch is aimed squarely at professionals who are not machine-learning engineers. Instead of hiring an expensive research team or consultants, lawyers, doctors and finance specialists can feed performance data back into a model and watch it get better at their own kind of work. The company says the platform is now publicly available, widening its market beyond AI specialists.
Early customers show what that looks like in practice. Legal AI company Harvey and financial analysis firm Rogo already use Trajectory to post-train open-source models, turning general models into assistants that behave more like experienced lawyers or analysts by learning from previous tasks — reviewing a particular type of legal contract, or valuing a company using a specific methodology. Decagon, Mercor and Nvidia are also named as customers.
A crowded race to automate training
Trajectory is not alone in chasing this bottleneck. It names Thinking Machines Lab, founded by former OpenAI chief technology officer Mira Murati and maker of the fine-tuning platform Tinker, as a direct competitor, alongside Applied Compute, which was valued at $3.25 billion last month, and Prime Intellect, which helps companies build and train their own AI agents.
Co-founder Michael Elabd argues that keeping people in the loop is also a safety feature: human involvement in the improvement process guards against autonomous improvement producing unintended or dangerous outcomes. For now, though, the company's simpler claim is the one investors paid for — that the next big gain in AI will come not from bigger models, but from models that quietly get better at one job by doing it every day.
