Trajectory raises $45M for continual-learning AI tools

Trajectory, an 18-person startup that three Stanford classmates founded roughly a year ago after working at Google DeepMind and Apple, has closed a $45 million Series A at a $300 million valuation, led by Sequoia, with Nvidia, Bessemer Venture Partners, and Bain Capital Ventures also participating.

The startup automates a process businesses otherwise handle with internal research teams or costly outside consultants: fine-tuning general AI models for specialized fields like law and finance. Its platform connects to a company’s existing applications, gathers feedback on how an open-source model performs, and retrains it so it keeps improving at a specific task — an approach Trajectory calls continual learning. Co-founder Ronak Malde likens off-the-shelf models to a brilliant PhD student perpetually starting a new job: “what the economy values is someone who worked on the job for 30 years,” he told TechCrunch.

Law firm-focused Harvey and financial-analysis startup Rogo already use the platform to post-train open-source models on tasks like reviewing specific contract types or valuing companies by a given methodology; Decagon, Mercor, and Nvidia are also customers. Trajectory is now opening the platform to the public. Michael Elabd, Trajectory’s other co-founder, argues that keeping humans in the loop — rather than letting a model improve itself autonomously — is what keeps the approach safe.

Trajectory also faces competition from Thinking Machines Lab — the fine-tuning startup former OpenAI CTO Mira Murati founded, known for its Tinker platform — as well as Prime Intellect and Applied Compute, the latter valued last month at $3.25 billion.

Why it matters

Trajectory’s bet is that most businesses don’t need a model that improves itself — they need one that improves with expert oversight, a distinction that matters as AI labs and regulators weigh the risks of fully autonomous self-improvement.