Trajectory, a startup founded by former researchers from Google DeepMind, OpenAI, and Apple, has launched a self-serve platform for post-training AI models. The system utilizes a lightweight software development kit to capture user edits and corrections to AI-generated outputs in real-time.
These captured signals are then used to feed post-training cycles, which can be executed on a weekly basis or more frequently. The platform includes a feature called continual grading to scale human preferences and the GS1 algorithm, which enables the model to learn directly from user actions.
The company, which exited stealth mode in May 2026, has secured approximately $60-65 million in total funding. A round led by Sequoia Capital valued the startup at $300 million. The founders of the venture are Ronak Malde, Arjun Karanam, and Michael Elabd.
To address enterprise concerns regarding privacy and data governance, Trajectory adheres to SOC 2 compliance standards. The platform is designed to allow companies to refine AI models for specific tasks, potentially outperforming broader models by learning from production data.
