Current AI models, whether developed by OpenAI or Anthropic or built on open source alternative models, have significant limitations. That means you won’t learn from your mistakes without additional training.
Companies that adapt these models to specialized fields, such as law or finance, typically work around this problem by tweaking the general model. However, this often requires hiring an in-house research team or relying on expensive consultants, commonly known as forward deployment engineers (FDEs).
Trajectory, a startup founded about a year ago by three Stanford classmates with experience at Google DeepMind and Apple, wants to automate that process, allowing AI models to learn from feedback and continually improve on specialized tasks.
The 18-employee company just announced that it has raised a $45 million Series A round at a valuation of $300 million, led by Sequoia and with participation from Nvidia, Bessemer Venture Partners, and Bain Capital Ventures.
Ronak Malde, co-founder of Trajectory, compared the foundational model to a very smart PhD student who is constantly starting new work. “But what the economy values is someone who’s been at the job for 30 years,” he told TechCrunch.
The startup aims to solve this problem through an approach known as continuous learning. Trajectory enables companies to start with open source AI models, connect the platform to their applications, and collect feedback on the model’s performance. Trajectory then uses that information to retrain the model, allowing it to improve on specialized tasks over time.
Trajectory allows lawyers, doctors and other professionals to almost automatically improve the models themselves, Marde said. This capability could make specialized AI deployment faster and cheaper.
Companies like Harvey and Rogo are already using Trajectory’s platform to post-train open source models, enabling them to function like experienced lawyers and financial analysts by learning from previous tasks, such as reviewing certain types of legal contracts or valuing companies using certain methodologies. Trajectory’s other customers include Decagon, Mercor, and Nvidia. The startup is currently rolling out its post-training platform to the public.
People using Trajectory can “decide what’s good for their organization and translate that into model improvements,” Malde says.
Trajectory’s approach to continuous learning has some similarities to recursive self-improvement (RSI), the idea that AI systems can improve their own abilities. However, co-founder Michael Elabud argues that keeping humans involved in the training process makes Trajectory’s approach safer, as autonomous self-improvement can lead to unintended or dangerous consequences.
Trajectory isn’t the only startup trying to automate post-training. Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, and the company behind direct competitor tweaking platform Tinker are also considering Applied Compute, which was valued at $3.25 billion last month, and Prime Intellect, which helps companies build and train their own AI agents.
Malde believes the startup’s biggest strength is its team. He was one of several researchers acquired by Google DeepMind from Windsurf. Elabd also worked at DeepMind, while Trajectory’s third co-founder, Arjun Karanam, previously worked on generative foundation models at Apple. The startup also hired researchers and product designers from OpenAI, Stripe, and Netflix.
Why is design so important for companies focused on AI training? Karanam said the startup has developed an interface that makes post-training more accessible to non-technical customers.
If you make a purchase through links in our articles, we may earn a small commission. This does not affect editorial independence.
