Trajectory opens its AI retraining platform to everyone with a $45 million round

Trajectory, a post-training software startup, said on November 2 that it is launching its platform publicly and has raised a $45 million Series A led by Sequoia at a $300 million valuation, according to its launch post. Nvidia, Bessemer Venture Partners and Bain Capital Ventures took part.

Three Stanford classmates and a year of work

TechCrunch reported that the 18-person company was founded about a year ago by three Stanford classmates. Co-founder Ronak Malde was among the researchers Google DeepMind hired from Windsurf, Michael Elabd also worked at DeepMind, and Arjun Karanam previously worked on generative foundation models at Apple.

Trajectory's launch post says its research team contributed to Gemini 3 at DeepMind, GPT-4o at OpenAI and infrastructure at Meta Superintelligence, and that its product team built interfaces at Stripe, Netflix and Apple.

What it sells: retraining a model on a company's own work

Post-training means adjusting an existing AI model after it is built, so it performs a specific job better. Trajectory's pitch is that a company can start from an open source model, connect it to its applications, score the results and let the platform retrain it. TechCrunch quoted Malde comparing general 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."

The company calls the platform "the first self-serve product for post-training on production data." It says one product replaces four (sandboxes, evals, training and deployment) and takes about five lines of code. Two features, Continual Grading and Group Size 1, are labeled research previews.

Who uses it, and what the company claims

Trajectory says builders at Harvey, Rogo, Mercor and Nvidia are already training on the platform, and TechCrunch adds Decagon to the list. The launch post says one customer trained more than eight models for long-horizon agent tasks in two weeks.

It also reports a result with Nvidia and Harvey: it took a Nemotron Lightning model "from 0% to an 8.3% all-pass rate on Harvey's LAB benchmark," a score it says matches Opus 5.5 at 100x lower cost and 3.8x faster speed. These are the company's own figures; the launch post does not publish revenue, and TechCrunch says none was disclosed.

A crowded field

TechCrunch lists other companies working on post-training, including Thinking Machines Lab, Applied Compute, which it says was valued at $3.25 billion last month, and Prime Intellect.

Customers named, revenue not yet

Our reading is that Trajectory has what the section looks for first: a launched product, named customers and a priced round from named investors, built by a team of 18 in a year. What it has not shown is revenue or how many customers pay, so the $300 million valuation is a milestone, not proof.

The threat we watch is dependence: the product trains other companies' models, and its best-known customers are a few large buyers. The opportunity is that a self-serve product could reach many smaller companies without the consulting work TechCrunch says rivals replace. We expect the next milestone to be a usage or revenue figure; if the company publishes one, it would show whether the Harvey result generalizes beyond a handful of logos.

The view in this final section is FinanceInTime's editorial opinion, for information only. It is not investment, tax, legal or other advice, and it does not take your circumstances into account. Forecasts are uncertain and past results do not guarantee future ones.