Trajectory raises $45M Series A for a self-serve post-training platform

Trajectory raises $45M Series A for a self-serve post-training platform

2 November 2026: Trajectory, which helps companies train AI models on their own production data, says it has raised a $45 million Series A led by Sequoia at a $300 million valuation. NVIDIA, Bessemer and Bain Capital Ventures took part. The company announced the round on 2 November together with the opening of its platform.

Not yet independently verified. Trajectory's post does not say where the company is based or what the money is for, and its home page still offers a waitlist although the post calls the platform self-serve. The benchmark result is the company's own claim. We will update this when it can be confirmed, and remove this note.

TechCrunch reports that the company was founded about a year ago by three Stanford classmates, Ronak Malde, Michael Elabd and Arjun Karanam, and has 18 people.

Post-training in five lines of code

Trajectory's platform covers training, evaluating and deploying a model in one place, and the company says a team can start with "about five lines of code". The pitch is that the people who know what a good answer looks like in law, finance or recruiting should be able to shape a model without a research team. Two research previews launched with it:

• Continual Grading, which learns how to grade outputs from a few examples of an expert's judgement;
• Group Size 1, which trains on live production interactions one at a time rather than on large batches prepared in advance.

"But what the economy values is someone who worked on the job for 30 years," Malde told TechCrunch, arguing that general models lack that judgement until it is trained in.

Customers and a benchmark claim

Trajectory names Harvey, Rogo, Mercor and NVIDIA as users; TechCrunch adds Decagon. The company says it took NVIDIA's Nemotron Lightning model from a 0% to an 8.3% all-pass rate on Harvey's LAB legal benchmark, matching a frontier model's score at a hundredth of the cost and 3.8 times faster. That result is Trajectory's own and has not been reproduced by anyone else.

TechCrunch names Thinking Machines Lab's Tinker, Applied Compute and Prime Intellect as companies selling into the same post-training market.

Where buyers meet it

Post-training platforms sit between model vendors and the teams that build on them. Our catalogue lists 7 machine learning software products and 143 models on its AI model board.

Company profile on TrustList: Trajectory

Sources

• Trajectory: Own Your Value Function, Introducing the Trajectory Platform, 2 November 2026
• TechCrunch: Trajectory raises $45M to make AI model training accessible to nontechnical experts, 2 November 2026