One-year-old startup Trajectory has announced the completion of a $45 million Series A funding round at a $300 million valuation. The round was led by venture capital firm Sequoia Capital, with participation from Nvidia, Bessemer Venture Partners, and Bain Capital Ventures.
The fundraising move comes as organizations seek to optimize AI operating costs and harness real-world efficiency as businesses begin to see financial benefits from AI.
Solving the post-training challenge for domain experts
Trajectory was co-founded by three Stanford alumni who previously worked at Google DeepMind and Apple: Ronak Malde, Michael Elabd, and Arjun Karanam. The team’s core mission is to solve the biggest limitation of current foundation models: they do not learn from their mistakes during real-world operations without continuous retraining.
Trajectory’s platform enables enterprises to connect open-source AI models directly into their workflows to collect user feedback. The system automatically aggregates this data to retrain models, helping the AI continuously adapt and improve execution quality for domain-specific tasks.
The breakthrough lies in its interactive interface: domain experts such as lawyers or doctors can directly steer and fine-tune models without requiring dedicated deployment engineering teams.
Early customer base and market competition
The company currently operates lean with 18 employees but has already attracted several prominent customers to its platform, including legal tech platform Harvey, financial startup Rogo, AI agent company Decagon, Mercor, and Nvidia itself. Alongside the funding announcement, Trajectory has officially opened access to its post-training platform to the broader market.
The automated post-training market is heating up with the emergence of formidable competitors such as former OpenAI Chief Technology Officer Mira Murati’s Thinking Machines Lab, Applied Compute, and Prime Intellect.
According to the founders, keeping humans in the loop within the continuous learning cycle is a far safer approach than allowing fully autonomous, unchecked recursive AI self-improvement.