Trajectory Raises $45 Million to Bring AI Model Training to Domain Experts

Trajectory has raised $45 million in a Series A round to help people without deep technical backgrounds improve AI models for specialized work. Sequoia led the round, with Nvidia, Bessemer Venture Partners and Bain Capital Ventures participating. The company was valued at $300 million.

The funding backs a pitch that AI model improvement should not be limited to engineering teams: specialists who understand a field’s day-to-day work should be able to help shape models for it. Trajectory says its platform uses continual learning to gather feedback from applications and retrain open-source models for specific tasks.

What Trajectory's Series A Means

Trajectory is targeting organizations that want AI tailored to professional workflows, rather than relying solely on a general-purpose model. The company points to lawyers and doctors as examples of specialists who could help improve models without taking on the technical work of training them directly.

That distinction matters because a model that performs well on general questions may still miss the terminology, priorities or expected outputs of a particular job. A legal team, for instance, may need assistance with a defined document-review task, while a healthcare organization may want a model adapted to a specific workflow. In each case, useful improvement depends on knowing what counts as a good answer in context.

Trajectory’s approach aims to make that expertise part of the improvement process. The Series A is a bet that organizations will pay for a simpler way to turn specialist feedback into more useful, task-specific AI - and that domain experts can contribute without needing to become machine-learning engineers.

How Its Continual-Learning Approach Works

The process described by Trajectory starts with a customer connecting an open-source model to an application. As people use the application, the company gathers feedback on how the model performs. That feedback is then used to retrain the model, with the goal of improving results on the customer’s specialized tasks.

In practice, this puts a feedback loop at the center of Trajectory AI training: use a model in a real workflow, identify where its responses fall short, and use those signals to improve its performance. The intended advantage is that a model can be adapted as an organization learns more about what it needs, rather than treating deployment as a one-time choice.

The approach also changes who may be able to participate. A specialist can judge whether an answer is relevant or useful, even if they cannot build a training pipeline themselves. The company’s proposition is that its tooling can help translate that judgment into model improvement with less direct technical involvement.

Why Specialized AI Training Matters - and What Organizations Still Need to Manage

Specialized AI models can be more useful when they are aligned with a narrow task, but adapting a model does not remove the work of deploying it responsibly. Open-weight models can give organizations more control over how they run and tailor AI. They also leave organizations with operational responsibilities, which can include managing infrastructure, monitoring performance and maintaining the system over time.

Teams considering this approach still need to decide what feedback is reliable, how model quality will be measured and who approves changes. For professional applications, they may also need safeguards for sensitive information and a clear process for checking outputs. These are important considerations whether retraining is handled internally or through a platform.

Trajectory’s funding highlights growing interest in making specialized AI training more accessible. Its proposed benefit is faster, less technically demanding adaptation of models for real work. Organizations weighing that promise should also assess the cost and responsibility of running and maintaining open-weight models before selecting a deployment approach.