Trajectory raises $45M Series A to automate AI model post-training
Trajectory, a startup founded about a year ago by three Stanford classmates with experience at Google DeepMind and Apple, announced that it raised $45 million in a Series A round at a $300 million valuation. Sequoia led the round, with participation from Nvidia, Bessemer Venture Partners and Bain Capital Ventures. The company has 18 people.
The pitch starts from a limitation of current AI models, whether built by OpenAI or Anthropic or based on open source alternatives: they do not learn from their mistakes without further training. Businesses adapting models for specialized fields such as law and finance usually get around this by fine-tuning general models. That often requires hiring internal research teams or relying on expensive consultants known as forward-deployed engineers (FDEs). Trajectory wants to automate that process, so models can learn from feedback and keep improving at specialized tasks.
Co-founder Ronak Malde compared foundation models to a very smart PhD student who is always starting a new job. "But what the economy values is someone who worked on the job for 30 years," he told TechCrunch.
The approach is called continual learning. A business starts with an open source model and connects Trajectory's platform to its applications. The platform collects feedback on how the model performs, then uses that information to retrain the model, so it gets better at specialized tasks over time. Malde said lawyers, doctors and other specialists can improve the models near automatically themselves, and that this could make deploying specialized AI faster and cheaper. In his words, the people using Trajectory decide what good looks like for their organization and can turn that into improvements for their model.
Harvey and Rogo already use the platform to post-train open source models. The aim is for the models to perform more like experienced lawyers or financial analysts by learning from previous tasks, such as reviewing a particular type of legal contract or valuing a company with a specific methodology. Other customers include Decagon, Mercor and Nvidia. The startup is now making its post-training platform publicly available.
The article notes that Trajectory's continual learning shares some similarities with recursive self-improvement (RSI), the idea that AI systems can improve their own capabilities. Co-founder Michael Elabd argues that keeping humans in the training process makes Trajectory's approach safer, since autonomous self-improvement could lead to unintended or even dangerous outcomes.
Trajectory is not the only startup trying to automate post-training. It names Thinking Machines Lab, founded by former OpenAI CTO Mira Murati and the company behind the fine-tuning platform Tinker, as a direct competitor. Others it lists are 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 advantage is its team. He was among the handful of researchers Google DeepMind acqui-hired from Windsurf. Elabd also worked at DeepMind, and the third co-founder, Arjun Karanam, previously worked on generative foundation models at Apple. The startup has also recruited researchers and product designers from OpenAI, Stripe and Netflix. Design matters, the article explains, because Karanam says the team built an interface that makes post-training more accessible to nontechnical customers.
Key facts
- Trajectory raised a $45 million Series A at a $300 million valuation, led by Sequoia, with Nvidia, Bessemer Venture Partners and Bain Capital Ventures participating.
- The 18-person startup, founded about a year ago by three Stanford classmates, automates post-training: it collects feedback on how an open source model performs and uses it to retrain the model.
- Harvey and Rogo already use the platform; Decagon, Mercor and Nvidia are also customers. The platform is now publicly available.
- Named competitors include Thinking Machines Lab (maker of Tinker), Applied Compute (valued at $3.25 billion last month) and Prime Intellect.
- Co-founder Michael Elabd argues that keeping humans in the training loop makes the approach safer than autonomous self-improvement.
Why it matters
Current models do not learn from their mistakes without further training, and the usual fix is fine-tuning, which often means an internal research team or expensive forward-deployed engineers. Trajectory wants to automate that step so models improve from feedback on their own tasks. If it works, the article says, deploying specialized AI could become faster and cheaper. The round also shows investor appetite for this niche: Trajectory's $300 million valuation sits beside a competitor, Applied Compute, valued at $3.25 billion last month.
Who it affects
Businesses that adapt AI models for specialized fields such as law and finance are the target, along with the specialists inside them. Malde says lawyers, doctors and other specialists can improve the models near automatically themselves. Harvey and Rogo are named users, and Decagon, Mercor and Nvidia are other customers. Teams that now depend on in-house researchers or FDE consultants for fine-tuning are the ones whose workflow this would change.
How to use it
The workflow described: start with an open source AI model, connect Trajectory's platform to your applications, let it collect feedback on how the model performs, and have it retrain the model for your specialized task. The people using it decide what good looks like for their organization. The post-training platform is now publicly available. Karanam says the company built an interface to make post-training accessible to nontechnical customers. No pricing is given.
How solid is it
The funding figures ($45 million, $300 million valuation, Sequoia as lead, the named participants) are stated plainly by the article. The product claims come from the founders and the company: no benchmark or measured result shows how much models improve with Trajectory, and no technical detail is given on how the retraining works. The article does not say what Harvey, Rogo or the other customers pay or what results they got. No revenue or customer count is given.
Risks and caveats
The article itself notes that Trajectory's continual learning shares some similarities with recursive self-improvement, and that autonomous self-improvement could lead to unintended or even dangerous outcomes. Elabd's answer is to keep humans in the training process, but that is his own argument; the article does not independently assess whether the approach is safer. Competition is crowded: Thinking Machines Lab, Applied Compute and Prime Intellect are all named, and Trajectory itself calls Thinking Machines Lab a direct competitor. Malde names the team as the startup's biggest advantage, which is a claim about people rather than a measured edge.
“But what the economy values is someone who worked on the job for 30 years”
— Ronak Malde, Trajectory co-founder, to TechCrunch