Trajectory Secures $45M Series A to Automate Model Training

Artificial intelligence startup Trajectory secured $45 million in a Series A round at a $300 million valuation to automate post‑training for specialized AI models (TechCrunch). The financing was led by Sequoia and included participation from Nvidia, Bessemer Venture Partners and Bain Capital Ventures.

The 18‑person company, founded about a year ago by three Stanford classmates with stints at Google DeepMind and Apple, has now opened its post‑training platform to the public. Co‑founder Ronak Malde highlighted early technical staff on X (Ronak Malde on X).

Trajectory targets a core limitation of foundation models from firms such as OpenAI or Anthropic: they do not learn from mistakes without ongoing training. Adapting these models for sectors like law and finance usually requires dedicated research teams or expensive engineers.

The platform uses a continual‑learning loop that plugs directly into customer applications. Companies start with open‑source models, collect performance feedback, and let Trajectory automatically retrain the models with domain‑specific guidance.

This setup lets subject‑matter experts—attorneys, physicians, and the like—refine models themselves instead of relying on machine‑learning engineers. Co‑founder Ronak Malde said users define what “good” looks like for their organization and translate that into model improvements.

Early adopters such as legal AI startup Harvey and financial platform Rogo are using the software to fine‑tune open‑source systems for tasks like contract review and financial modeling. Reported enterprise customers also include support firm Decagon, talent marketplace Mercor, and chipmaker Nvidia.

Trajectory’s workflow echoes ideas of recursive self‑improvement, but co‑founder Michael Elabd stressed that human oversight prevents unintended or hazardous outcomes that fully autonomous loops might cause. The company competes with rivals including Applied Compute, Prime Intellect and Mira Murati’s Thinking Machines Lab.

The investor interest in Trajectory suggests that the enterprise artificial intelligence market is shifting focus from basic model access to operational refinement. By allowing domain experts rather than specialized machine learning engineers to direct post-training, the platform appears likely to reduce deployment expenses for regulated industries. However, competing against well-funded rivals like Applied Compute and Thinking Machines Lab will likely test whether Trajectory's continuous feedback mechanism can maintain model safety and performance at scale.

Spotted an error? See our corrections policy or email editorial@lordofclaude.com.