Trajectory, a startup founded by former Google DeepMind and Apple employees, has raised $45 million in a Series A funding round at a $300 million valuation. The round was led by Sequoia with participation from Nvidia, Bessemer Venture Partners, and Bain Capital Ventures, the company reported.
The 18-person startup aims to automate the process of fine-tuning AI models for specialized fields like law and finance. Currently, businesses often rely on expensive consultants or internal research teams to adapt general models to specific tasks. Trajectory utilizes a method known as continual learning, which allows an open source model to collect feedback from applications and use that data to retrain itself and improve over time.
Co-founder Ronak Malde compared foundational models to a highly intelligent PhD student who is constantly starting a new job. “But what the economy values is someone who worked on the job for 30 years,” Malde said.
By providing an interface designed for nontechnical users, the company intends to allow specialists such as doctors and lawyers to improve models near automatically. This approach aims to make the deployment of specialized AI faster and cheaper. Current customers using the platform to post-train open source models include Harvey, Rogo, Decagon, Mercor, and Nvidia.
While the company's approach shares similarities with recursive self-improvement, co-founder Michael Elabd argued that keeping humans involved in the training process makes the method safer than autonomous self-improvement. The startup faces competition from several other firms, including Thinking Machines Lab, Applied Compute, and Prime Intellect.
The founding team includes Malde, who was part of a group of researchers Google DeepMind acquired from Windsurf, Elabd, a former DeepMind employee, and Arjun Karanam, who previously worked on generative foundation models at Apple. The company has also recruited staff from OpenAI, Stripe, and Netflix.
