Startup Trajectory raises $45 million to simplify AI model training
Trajectory, a US startup that automates post-training of AI models, has raised $45 million in a Series A round at a $300 million valuation. The round was led by Sequoia.
Trajectory, an AI startup building a platform for post-training of AI models, has announced a $45 million Series A funding round at a $300 million valuation. The round was led by venture firm Sequoia, with participation from chipmaker Nvidia, Bessemer Venture Partners and Bain Capital Ventures. The platform is meant for companies that want to adapt AI models to specialized fields such as law and finance, and the company says it can make that faster and cheaper.
The problem being addressed
Current AI models, whether built by OpenAI or Anthropic or based on open source alternatives, do not learn from their mistakes without further training. To adapt them to narrow fields, businesses usually fine-tune general models. That often requires in-house research teams or costly consultants known as forward-deployed engineers (FDEs).
Trajectory wants to automate this process through an approach called continual learning. A business starts with an open source model and connects the platform to its applications. The platform gathers feedback on how the model performs and uses it to retrain the model, improving it at specialized tasks over time. Co-founder Ronak Malde said lawyers, doctors and other specialists will be able to improve the models almost automatically themselves.
Customers and competitors
Harvey and Rogo already use the platform to make open source models perform more like experienced lawyers or financial analysts. Other customers include Decagon, Mercor and Nvidia. The platform is now being made publicly available.
Direct competitors include Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, as well as Applied Compute and Prime Intellect. Trajectory was founded about a year ago and has 18 employees. Its founders are three Stanford classmates with experience at Google DeepMind and Apple. Co-founder Michael Elabd argues that keeping humans involved in training makes the approach safer than fully autonomous self-improvement.
