Trajectory secures $15m seed for continuous AI learning platform
The new venture, backed by Conviction and individual investors including Jeff Dean, aims to help businesses refine AI products through real-time user feedback loops.
A cohort of artificial intelligence researchers from Google DeepMind, Apple, OpenAI, and Meta Superintelligence Labs has launched a new startup named Trajectory. The company is developing a platform designed to enable businesses to continuously improve their AI products by training on real-world user interactions, addressing a significant limitation in current generative AI systems.
Trajectory has raised a $15 million seed round at a $115 million post-money valuation. The round was led by the venture capital firm Conviction, with participation from Bessemer Venture Partners, Radical VC, and BoxGroup. Individual investors in the funding include Google DeepMind chief scientist Jeff Dean and Stanford professor Fei-Fei Li.
The startup’s leadership includes CEO and co-founder Ronak Malde, who previously worked at Windsurf before joining Google DeepMind following its acquisition. Co-founders Arjun Karanam, a former Apple researcher on the Vision Pro, and Michael Elabd, formerly of Google DeepMind’s robotics division, are also part of the team. The company currently employs 11 researchers and engineers.
The core technology involves logging instances where AI systems fail or require human intervention, such as customer support queries that are bounced to human agents. These data points are used to post-train open-source models on specific business needs. Trajectory claims these post-trained models outperform frontier lab models on narrow tasks relevant to specific products.
Current customers include enterprise sales firm Clay and legal AI startup Harvey. The company aims to reduce the reliance on forward-deployed engineers by building a product that improves autonomously. While currently focused on AI-native companies, Trajectory plans to market its platform to the Fortune 500 in the future.
Co-founder Michael Elabd stated that the eventual goal is to update models daily, hourly, or per interaction. This approach aligns with arguments made by Turing award winner Richard Sutton at the December 2025 NeurIPS conference, which posited that continual learning is essential for building superintelligent agents.
The announcement was made on a Wednesday, highlighting the startup’s ambition to replicate the rapid iteration cycles seen in AI coding tools across broader industries. Trajectory seeks to provide the infrastructure for companies to move beyond static AI deployments that do not adapt to new data or errors.
