Trajectory Raises $45M to Make AI Training Accessible to Nontechnical Experts

Trajectory, a 18-person AI startup founded about a year ago by Stanford graduates with prior experience at Google DeepMind and Apple, has raised $45 million in a Series A round at a $300 million valuation. The financing was led by Sequoia Capital and included participation from Nvidia, Bessemer Venture Partners, and Bain Capital Ventures. The funding will support Trajectory’s effort to make advanced model post-training accessible to nontechnical experts, enabling organizations to customize AI systems without relying on costly external engineering teams. The company is building a platform for continual learning, a method that allows AI models to improve at specialized tasks by learning from real-world feedback. Foundational models from companies such as OpenAI and Anthropic, as well as open-source alternatives, can perform broad tasks well but do not inherently learn from their mistakes once deployed. To make them useful in fields such as law, finance, or medicine, businesses typically fine-tune models using internal researchers or forward-deployed engineers. Trajectory aims to reduce that dependency by automating much of the post-training workflow. Trajectory’s platform lets businesses begin with an open-source AI model, connect it to their existing applications, and gather data on how the model performs in real use cases. The system then uses that feedback to retrain the model, helping it improve over time. Co-founder Ronak Malde said the goal is to allow lawyers, doctors, and other domain experts to refine models themselves, making specialized AI deployment faster and cheaper. Early customers already include Harvey, Rogo, Decagon, Mercor, and Nvidia. The platform has been used to help open-source models perform more like experienced legal reviewers or financial analysts by learning from repeated tasks, such as reviewing specific types of contracts or applying defined valuation methods. Trajectory is now making its post-training platform publicly available. Co-founder Michael Elabd said Trajectory’s approach differs from recursive self-improvement, a concept in which AI systems autonomously enhance their own capabilities. He argued that keeping humans involved in defining quality and guiding feedback makes the process safer by reducing the risk of unintended or potentially harmful outcomes. The startup counts Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, along with Applied Compute and Prime Intellect, as competitors in the growing field of automated model customization and AI agent development. Trajectory believes its advantage lies in its team and product design. Its founders include Malde, an early Google DeepMind researcher from the Windsurf acqui-hire, Elabd, also a former DeepMind researcher, and Arjun Karanam, who previously worked on generative foundation models at Apple. The company has also hired talent from OpenAI, Stripe, and Netflix. Karanam said Trajectory has focused heavily on creating an interface that allows non-technical customers to control how models improve according to their organization’s standards.