Trajectory Raises $45 Million at a $300 Million Valuation to Automate AI Post-Training and Continual Learning

Whether developed by OpenAI, Anthropic, or built on top of various open-source alternatives, current artificial intelligence models share a fundamental and critical limitation: they do not natively learn from their own mistakes without further training and human intervention.

For businesses attempting to adapt these general-purpose foundation models for specialized and highly regulated fields—such as law, medicine, and finance—this constraint typically forces organizations to rely on costly workarounds. Companies usually get around this barrier by fine-tuning general models to suit their specific workflows. However, executing this effectively often requires hiring expensive internal research teams or depending heavily on costly external consultants, a class of professionals commonly known in the industry as forward-deployed engineers, or FDEs.

Trajectory, an artificial intelligence startup founded roughly a year ago by three Stanford classmates with deep technical experience at Google DeepMind and Apple, is setting out to entirely automate that process. The company’s platform is designed to allow AI models to learn continuously from feedback and steadily improve at specialized, domain-specific tasks without requiring an army of machine learning engineers.

Demonstrating strong investor appetite for solutions that bridge the gap between general AI and enterprise utility, the 18-person company just announced that it has successfully raised $45 million in a Series A funding round. The financing values the startup at $300 million and was led by venture capital powerhouse Sequoia, with significant participation from prominent backers including Nvidia, Bessemer Venture Partners, and Bain Capital Ventures.

To explain the core problem Trajectory is solving, co-founder Ronak Malde offered a vivid analogy, comparing today’s foundational AI models to a brilliantly talented PhD student who is nevertheless perpetually starting their very first day on a new job.

"But what the economy values is someone who worked on the job for 30 years," Malde told TechCrunch, highlighting the vast performance gap between raw, general intelligence and accumulated domain expertise.

The startup aims to solve this persistent problem through an established machine learning paradigm known as continual learning. Businesses that utilize Trajectory can start with an existing open-source AI model and connect the platform directly to their enterprise applications. This integration allows the system to seamlessly collect feedback on how the model performs in real-world scenarios. Trajectory then takes that incoming information and systematically retrains the model, helping it adapt, evolve, and dramatically improve at specialized tasks over time rather than remaining static.

With Trajectory’s technology integrated into their workflows, professionals such as lawyers, doctors, and other domain specialists can improve underlying models almost automatically and entirely by themselves, Malde explained. This operational capability has the potential to make the deployment of specialized enterprise AI significantly faster, more efficient, and much less expensive for organizations of all sizes.

Early traction suggests a strong market appetite for this approach. Companies like Harvey and Rogo are already utilizing Trajectory’s platform to post-train open-source models. By doing so, these firms help their AI systems perform more like experienced lawyers or veteran financial analysts, learning directly from previous tasks—such as reviewing a specific and nuanced type of legal contract or valuing a corporate entity using a highly specific methodology. Trajectory’s growing roster of enterprise customers also includes prominent industry players like Decagon, Mercor, and Nvidia. Following this latest infusion of venture capital, the startup is now making its advanced post-training platform publicly available to a wider audience.

When using Trajectory, the human operators on the ground decide "what good looks like for their organization and can turn that into improvements for their model," Malde noted, emphasizing that human expertise remains at the center of the improvement loop.

Interestingly, Trajectory’s approach to continual learning shares some high-level conceptual similarities with recursive self-improvement, commonly known as RSI—the theoretical idea that advanced AI systems can autonomously improve their own capabilities over time without human guidance. However, co-founder Michael Elabd argues that keeping humans actively involved in the training process makes Trajectory’s specific approach considerably safer. He points out that fully autonomous, unsupervised self-improvement could easily lead to unintended, unpredictable, or potentially dangerous operational outcomes for enterprise environments.

Despite its unique positioning, Trajectory is far from the only startup attempting to automate the complex and resource-intensive market for post-training and model refinement. Other ambitious companies going after this lucrative market include Thinking Machines Lab, a venture founded by former OpenAI CTO Mira Murati, which stands behind the fine-tuning platform Tinker. Additional competitors include Applied Compute, which commanded a staggering valuation of $3.25 billion last month, and Prime Intellect, a platform dedicated to helping enterprises build and train their own autonomous AI agents.

When asked about how the startup plans to stand out in an increasingly crowded competitive landscape, Malde expressed confidence that the team’s collective pedigree is its biggest strategic advantage. Malde himself was among the small handful of elite researchers that Google DeepMind acqui-hired from Windsurf. Elabd also spent time working at DeepMind, while Trajectory’s third co-founder, Arjun Karanam, previously worked extensively on generative foundation models at Apple. Beyond the founding trio, the startup has successfully recruited top-tier researchers and product designers from industry heavyweights such as OpenAI, Stripe, and Netflix.

Given that the company’s core product revolves around advanced machine learning infrastructure, questions naturally arise as to why design and user experience matter so much to a firm focused primarily on AI training. Karanam emphasizes that the startup has deliberately developed an intuitive interface designed to make post-training accessible and straightforward for nontechnical enterprise customers, ensuring that domain experts do not need a background in computer science to make their AI models smarter.