Trajectory Secures $45 Million to Let Nontechnical Experts Teach and Refine AI

Trajectory says its tools let lawyers, clinicians and analysts improve models without hiring research teams.

Trajectory, a year-old startup formed by three Stanford alumni with stints at DeepMind and Apple, raised $45 million in a Series A round that values the business at about $300 million. The round was led by Sequoia and drew participation from Nvidia, Bessemer Venture Partners and Bain Capital Ventures. The company, which employs roughly 18 people, says its software helps organizations keep AI models current by collecting real-world feedback and using it to retrain those models for domain-specific tasks.

The problem Trajectory aims to address is simple: out-of-the-box AI systems don’t change their behavior unless someone retrains them. Firms that want models to act like seasoned specialists usually must hire internal research staff or pay outside engineers who embed with teams. Trajectory offers a different path. Customers plug an open-source model into the startup’s platform, attach it to their workflows, and let the system gather signals about where the model succeeds and where it falters. That information is then used to refine the model so it performs better on repeated, specialized assignments.

For end users — attorneys reviewing niche contracts, financial teams valuing companies under a specific framework, or medical professionals checking notes — the appeal is reduced time and cost to deploy customized AI. Trajectory says early adopters have been using the service to post-train models so they behave more like experienced practitioners. The company named a handful of customers that have tested the platform and announced it is making the post-training service generally available this month.

Keeping humans in the loop is central to Trajectory’s pitch on safety. While some approaches envision systems that iteratively and autonomously improve their own abilities, Trajectory’s founders argue that maintaining human oversight lowers the chance of unintended behavior from model updates. Independent researchers who study machine learning governance say gradual, supervised retraining with clear success metrics tends to be more predictable than purely self-directed optimization, though they add that careful auditing remains essential.

Trajectory enters a market where several startups are racing to simplify fine-tuning and post-training. Competing offerings include tools spun out by ex–OpenAI staff and other platforms that package continuous learning for enterprises; some rivals have recently commanded multibillion-dollar valuations. Trajectory’s leaders point to their personnel as a differentiator: many on the roster previously worked at DeepMind, OpenAI, Apple and major tech firms, and the company focused early on designing an interface meant to make iterative improvement accessible to nontechnical teams.