Trajectory raises $45 million to help AI teams learn from production feedback

Trajectory announced a $45 million Series A on November 2 at a $300 million valuation. Sequoia led the round, with participation from Nvidia, Bessemer Venture Partners and Bain Capital Ventures. The financing accompanied the launch of its platform for post-training AI models on production data.

What Trajectory is building

When a user corrects an AI assistant, the correction gives the team a clue. The model may have misunderstood the task, the application may have retrieved the wrong document, or the instructions may have been unclear. Each problem calls for a different fix. In its platform announcement, Trajectory describes a way to use feedback from real usage to improve models on specialized tasks. TechCrunch also reported the round and covered the company’s effort to make model training accessible beyond specialist research teams.

What buyers should look for

Consider a support assistant that repeatedly gives the wrong refund answer. If it reads an old policy, updating the source may solve the problem. If it consistently misapplies the current policy despite finding the right document, the team has a better candidate for model improvement. Either way, fresh test cases can show whether the fix works and whether it breaks answers that were previously correct. Three questions are worth asking before adopting a post-training workflow: Which feedback is reliable enough to train on? How will the team measure results on examples held out from training? Who approves or rolls back a model update? Customer results that answer those questions will tell buyers more about the platform than the size of the round.