Trajectory

Trajectory is a platform for continual learning that converts real-world product usage — edits, retries, accepts, and corrections — into ongoing AI model improvements. Built by researchers with backgrounds in continual learning for robotics, 3D foundation models, and SWE-1, the platform is backed by Conviction and AI pioneers including Fei-Fei Li and Jeff Dean. It targets AI product teams that want their models to get smarter from production usage rather than relying on static, frozen datasets.

What It Is

Trajectory sits in the category of AI training infrastructure and model lifecycle management. Its core job is to close the feedback loop between how users actually interact with an AI-powered product and how the underlying model behaves. Instead of periodic, manual retraining cycles, Trajectory enables a living system where user behavior continuously informs model updates — a paradigm the team calls "continual learning."

How the Platform Works

The platform is structured around four stages:

• Instrument — A lightweight SDK plugs into an existing product and begins capturing behavioral signals: traces, corrections, re-prompts, and edits that users generate naturally.
• Understand — The platform surfaces patterns in those signals so teams can see where the model is succeeding or failing in real usage.
• Steer — Teams direct the model toward desired behaviors, defining what matters and what the system should learn from.
• Learn — Model updates are generated from the curated signals and deployed after passing through the team's own eval suite and approval workflow.

Data Governance and Trust Model

Trajectory is SOC 2 certified and emphasizes customer control over training data. The platform states that users decide which data is used for training, that no model update deploys without passing the customer's own eval suite and approval workflow, and that every training run is fully auditable. The vendor describes this as "your data, trained and deployed on your terms."

Customer Signals

The Trajectory website attributes several customer quotes to named individuals. Kareem Amin, CEO of Clay, is quoted saying Trajectory is "building the infrastructure to help us explore" continual learning and that Clay has "already seen examples of it learning from its mistakes." Cyrus Asgari, Research Engineer at Decagon, describes a collaboration focused on measuring and improving model steerability in post-trained models. Gabe Pereyra, President and Co-Founder of Harvey, describes using Trajectory to build agents that carry "evolving domain expertise" for legal work.

Team and Backing

The team describes itself as researchers who pioneered continual learning for robotics, 3D foundation models, and SWE-1. The company is backed by Conviction and lists AI pioneers Fei-Fei Li and Jeff Dean as backers. The homepage also shows logos associated with Notion, Dropbox, and other organizations, though the nature of those relationships is not explicitly stated in the source text.