Trajectory, a startup founded about a year ago by three Stanford classmates with professional experience at Google DeepMind and Apple, aims to address a common limitation of today’s large AI models: they do not improve from mistakes without further training.
The 18-person company announced it raised $45 million in a Series A round at a $300 million valuation. The round was led by Sequoia, with participation from Nvidia, Bessemer Venture Partners and Bain Capital Ventures.
The problem and Trajectory’s approach
Current foundation models — whether from OpenAI or Anthropic or built on open-source alternatives — typically require fine-tuning to perform well on specialized tasks. Organizations often accomplish that by hiring in-house research teams or expensive consultants known as forward-deployed engineers (FDEs).
Trajectory’s platform seeks to automate that post-training process using an approach known as continual learning. Customers start with an open-source or general model, connect Trajectory to their applications, and the platform collects feedback about model performance. Trajectory then uses those signals to retrain the model so it improves over time on domain-specific tasks.
Co-founder Ronak Malde compared foundation models to a very smart PhD student who is always starting a new job, while what the economy often values is someone with decades of on-the-job experience. Trajectory says its system can make domain adaptation nearly automatic, enabling professionals such as lawyers or doctors to iteratively improve model behavior without building large research teams.
Customers, launch and availability
Companies including Harvey and Rogo already use Trajectory’s platform to post-train open-source models so they behave more like experienced lawyers or financial analysts — for example, when reviewing particular types of legal contracts or valuing companies using specific methodologies. Other customers include Decagon, Mercor and Nvidia. Trajectory is now making its post-training platform publicly available.
The company emphasizes that customers define what “good” looks like for their organization and can translate those standards into model improvements.
Safety, competitors and the team
Trajectory’s continual learning approach overlaps conceptually with recursive self-improvement (RSI), the notion that AI systems can iteratively improve their own capabilities. Co-founder Michael Elabd argues that retaining human involvement in the training loop makes Trajectory’s method safer, since autonomous self-improvement could produce unintended or dangerous outcomes.
Trajectory faces competition from several startups working on automated post-training. It names Thinking Machines Lab (founded by former OpenAI CTO Mira Murati) and the company behind the fine-tuning platform Tinker as direct competitors, and notes Applied Compute — which was valued at $3.25 billion last month — and Prime Intellect, which helps companies build and train their own AI agents, as other players in the space.
The company stresses team composition as a key advantage. Founders and employees include researchers previously at Google DeepMind (some acquired in the Windsurf deal), as well as engineers and designers from Apple, OpenAI, Stripe and Netflix. Co-founders Ronak Malde, Michael Elabd and Arjun Karanam bring relevant research and industry experience; Karanam previously worked on generative foundation models at Apple.
Why does product design matter for an AI training company? Arjun Karanam says Trajectory has developed an interface that makes post-training accessible to nontechnical customers, reducing the need to hire extensive internal research staff.
Why this matters
By automating post-training and lowering the barrier to continual learning, Trajectory’s platform could reduce costs and speed up the deployment of specialized AI solutions for industry users. At the same time, competition in the space and safety considerations around automated model improvement will determine how broadly and quickly these approaches are adopted.
