Trajectory has raised $45 million to make AI model training more accessible to businesses without specialist research teams. The startup says its platform is now publicly available, aiming to turn feedback on a model’s work into improvements for specialised tasks.
Watch Desk analysis
What happened
Trajectory’s Series A values the year-old company at $300 million, according to TechCrunch’s report. Sequoia led the round, with Nvidia, Bessemer Venture Partners and Bain Capital Ventures also participating.
The company’s approach is to connect an open-source AI model to a business’s applications, collect feedback on how it performs and use that information to retrain the model. Trajectory calls this continual learning. Its co-founder Ronak Malde told TechCrunch the aim is to let specialists decide what good work looks like and turn that into model improvements.
TechCrunch says Harvey and Rogo are using the platform to post-train open-source models for legal and financial work. Decagon, Mercor and Nvidia are also named as customers. Trajectory has 18 employees, the report says.
Why it matters
Fine-tuning a model for a particular business can mean hiring an internal research team or bringing in expensive consultants. Trajectory is pitching a different route: give specialists a way to feed work outcomes back into a model without asking them to become machine-learning engineers first.
That could make specialised AI easier to adapt, but the useful test is not whether feedback can be collected. It is whether the resulting model improves on real tasks, reliably enough to justify giving it a bigger role in the work.
Our read
The interesting promise here is not that a model improves itself in a mysterious corner. It is that a business can define its own standard and use feedback to refine a model over time. That is a meaningful step towards practical, tailored AI, particularly if the alternative is a costly training project.
But “accessible” needs to mean more than a friendlier interface. Businesses will want to know how much expertise and oversight the process still needs, what kinds of feedback work, and how they can tell a genuine improvement from a model that has simply learnt to please its reviewer. There is a product to try here, and a fair few questions to take along.
What to watch
- Whether Trajectory publishes concrete results from its customer deployments.
- How much technical work businesses still need to do to connect applications and train models.
- Whether customers can show that feedback improves performance on specialised tasks over time.
Discussion spark: Should businesses trust specialists to steer AI models through day-to-day feedback, or should model improvements remain the preserve of dedicated research teams?
Sources and evidence
- Trajectory raises $45M to make AI model training accessible to nontechnical experts (2 November 2026, 23:13 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.