Trajectory, an 18-person startup founded about a year ago by three Stanford classmates, has raised a $45 million Series A at a $300 million valuation. Sequoia led the round, and Nvidia, Bessemer Venture Partners and Bain Capital Ventures also put in money, according to TechCrunch AI. The company is also opening its post-training platform to the public. The pitch is simple: let lawyers, doctors and financial analysts improve AI models themselves, without hiring a research team.
The Short Version
- What: Trajectory automates continual learning. Models collect feedback from real use and get retrained to do better at specialized tasks.
- Who’s using it: Harvey, Rogo, Decagon, Mercor and Nvidia.
- Why it matters: Today, fine-tuning a model for law or finance usually means an in-house research team or expensive forward-deployed engineers (FDEs). Trajectory wants to make that a self-serve job for subject experts.
- What to watch: Post-training is getting crowded and expensive to compete in. Thinking Machines Lab and Applied Compute are both after the same market.
The Problem Trajectory Is Going After
Every major model has the same blind spot. GPT, Claude or an open source alternative won’t learn from its mistakes unless someone trains it again. If your AI misreads a certain kind of contract on Monday, it’ll misread it the same way on Friday.
Co-founder Ronak Malde describes a foundation model as a very smart PhD student who’s always starting a new job. “But what the economy values is someone who worked on the job for 30 years,” he told TechCrunch.
That gap is why so many enterprise AI rollouts stall. The model is smart enough. It just never builds up the experience that makes a senior employee worth paying.
How It Works
The workflow is easy to follow, even if the machinery underneath isn’t:
- A company starts with an open source model.
- It connects Trajectory to the apps where that model does its work.
- The platform collects feedback on how the model performs.
- Trajectory uses that feedback to retrain the model, so it keeps getting better at the task.
The important part is who decides what “better” means. Customers define “what good looks like for their organization and can turn that into improvements for their model,” Malde said. A legal team can tune a model on one specific type of contract review. A finance firm can teach it its own valuation method.
That’s also why design matters here. Co-founder Arjun Karanam says the team built an interface specifically so nontechnical customers can handle post-training on their own. The startup has hired product designers from Stripe and Netflix along with researchers from OpenAI.
Humans in the Loop, on Purpose
Continual learning sounds a lot like recursive self-improvement (RSI), the idea that AI systems can upgrade their own abilities. Co-founder Michael Elabd says the difference is deliberate. Trajectory keeps humans in the training process, which he argues makes it safer than letting a system improve itself on its own, where the results could be unintended or even dangerous.
What stands out to me is that this is a safety argument and a sales argument at once. Enterprises in regulated fields like law and medicine aren’t going to hand training decisions over to a black box. Keeping experts in control is what lets those buyers say yes.
A Crowded, Well-Funded Field
Trajectory isn’t alone. The post-training race already includes:
- Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, which makes the fine-tuning platform Tinker.
- Applied Compute, valued at $3.25 billion just last month.
- Prime Intellect, a platform that helps companies build and train their own AI agents.
Malde says Trajectory’s edge is its team. He was one of the handful of researchers Google DeepMind acqui-hired from Windsurf. Elabd also worked at DeepMind, and Karanam worked on generative foundation models at Apple.
Why Practitioners Should Care
The bigger story is where value in AI is moving. Frontier labs compete on raw intelligence. More and more, the money is in turning that intelligence into expertise in a specific domain. Open source models plus easy post-training could give companies a real alternative to paying a premium for closed APIs, and they’d own the improvements.
If you’re deploying AI in a specialized field, a few things are worth doing now:
- Capture feedback now. Corrections, approvals and expert edits are training data. Start logging them.
- Check out open models. Platforms like this one assume you start from open weights.
- Rethink the FDE budget. If self-serve post-training works, a lot of consulting spend becomes optional.
With Harvey and Rogo already on board and the platform now public, Trajectory has a real shot at showing that AI can learn on the job. The full story is at TechCrunch AI.