Sequoia Leads $45M Series A in AI Training Startup Trajectory

Made possible with backing from

Trajectory, a startup founded about a year ago, has raised $45 million in a Series A round at a $300 million valuation. Sequoia led the round.

The gap it targets

Models built by OpenAI, Anthropic or the open source community all stumble on the same point: after release they only get better if someone trains them again.

Law and finance firms usually adapt a general model through fine-tuning. That often means in-house researchers or expensive consultants, known as forward-deployed engineers, or FDEs.

Trajectory wants to automate that work. Its aim is for models to learn from feedback and keep improving at specialized tasks.

How the platform works

The approach is called continual learning. A business starts with an open source model and connects the platform to its applications. The platform collects feedback on how the model performs, then retrains the model with it.

Co-founder Ronak Malde likened 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 said.

Malde said specialists such as lawyers and doctors can improve the models themselves, almost automatically. That could make deploying specialized AI faster and cheaper. Users decide what good looks like for their organization.

Customers, investors and rivals

Harvey and Rogo already use the platform to post-train open source models, so they behave more like experienced lawyers or financial analysts. Decagon, Mercor and Nvidia are also customers. Trajectory is now opening the post-training platform to the public.

Nvidia also took part in the round, as did Bessemer Venture Partners and Bain Capital Ventures.

The field is crowded. Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, and its fine-tuning platform Tinker compete here. So do Applied Compute, valued at $3.25 billion last month, and Prime Intellect.

A small team with big-lab roots

Three Stanford classmates started the company. Malde was among the researchers Google DeepMind acqui-hired from Windsurf. Michael Elabd also worked at DeepMind. Arjun Karanam previously worked on generative foundation models at Apple. The team has hired from OpenAI, Stripe and Netflix, and has 18 people in total.

Karanam says the company built an interface that makes post-training reachable for nontechnical customers.

Trajectory's approach resembles recursive self-improvement, where AI systems improve their own abilities. Co-founder Elabd argues that keeping humans in the training loop is safer, since a system left to improve itself without people might produce unintended or even dangerous results.

Our view

The pitch rests on a claim that domain experts, not engineers, can steer model quality. That is testable: the open question is whether customers beyond the named law and finance users get the same gains. The public release will show whether the interface works for nontechnical teams.

Source: TechCrunch