Trajectory, a startup founded about a year ago by three Stanford classmates, announced a $45 million Series A led by Sequoia on November 2, according to TechCrunch's report. The company says its platform lets businesses retrain open source AI models from feedback so the models keep improving at specialized work, and it is now making that platform publicly available.
TechCrunch reports a $300 million valuation, with Nvidia, Bessemer Venture Partners and Bain Capital Ventures also participating. For founders building products on top of AI models, the problem Trajectory targets deserves attention, because it asks how a company turns a general model into one that behaves like a specialist without hiring a research team.
This post covers what Trajectory is building, where it sits among competitors, and what founders in Palo Alto, Redwood City and across the Peninsula can take from it. Every fact comes from TechCrunch's reporting, and the numbers are attributed to that source.
What Trajectory is building
TechCrunch says current AI models share a limitation, which is that they do not learn from their mistakes without further training. Businesses usually get around this by fine-tuning general models, and that often requires an internal research team or consultants known as forward-deployed engineers. Trajectory wants to automate that work through continual learning. A company starts from an open source model, connects the platform to its applications, and lets the platform collect feedback on how the model performs and retrain it over time.
Co-founder Ronak Malde told TechCrunch that lawyers, doctors and other specialists can improve the models themselves, nearly automatically, by deciding what good looks like for their organization. TechCrunch reports that Harvey and Rogo already use the platform to post-train open source models, and that Decagon, Mercor and Nvidia are also customers.
The team behind it
The company has 18 people. Malde was among the researchers Google DeepMind acqui-hired from Windsurf, co-founder Michael Elabd also worked at DeepMind, and co-founder Arjun Karanam previously worked on generative foundation models at Apple. TechCrunch adds that the startup has recruited researchers and product designers from OpenAI, Stripe and Netflix.
Where it sits in a crowded field
TechCrunch names several competitors working on automated post-training. Thinking Machines Lab, which makes the fine-tuning platform Tinker, is described as a direct competitor, along with Applied Compute, which TechCrunch says was valued at $3.25 billion last month, and Prime Intellect, which helps companies build and train their own AI agents.
The competitors take different routes to a similar goal. The Tinker page showcases a customer that trained an open model on financial information-filtering tasks, while Prime Intellect's site describes training specialist models with supervised fine-tuning and then reinforcement learning against rewards you define from real outcomes. Trajectory's pitch sits at the easier end of that spectrum, since it aims to let nontechnical experts steer the process.
Malde told TechCrunch the startup's biggest advantage is its team, and Karanam says it built an interface that makes post-training accessible to nontechnical customers. Elabd also draws a line between Trajectory's approach and recursive self-improvement, arguing that keeping humans in the training loop is safer than autonomous self-improvement, which could lead to unintended outcomes.
What founders can take from it
Three practical ideas follow from the reporting, offered here as suggestions rather than predictions. First, if your product serves a specialist field, the corrections and approvals your own experts already produce are training signal, so capture them from your first customers. Second, post-training platforms may let a small team postpone hiring dedicated research engineers, although you should test any vendor against your own quality bar before you commit.
Third, ask vendors who defines what good looks like. Trajectory's pitch depends on customers setting that standard, which puts your domain experts at the center of the process.
Before you adopt any post-training service, write down a short checklist. Ask where your feedback data is stored and who can see it, how you measure whether a retrained model is better than the one you started with, and what happens to your tuned model if you switch providers. Those answers matter more once the model carries your customers' work. For more recent funding stories with a founder's lens, see our posts on Baseten's reported $30B round, &Truth's $70M raise and Serval's $150M Series C and its plan to hire new grads.
The Peninsula angle
Stanford classmates starting AI companies is a familiar Peninsula story, and the people who want to compare notes on it are close by. South Bay Renaissance, which Tribe organizes to rebuild the South Bay and Peninsula startup and social scene in person, hosted Inside Native AI Companies With Gokul Rajaram in Redwood City on October 21, a conversation about how AI-native companies structure teams, talent and agents.
Founders thinking about customers rather than models have another option. The Tribe Events calendar lists an Enterprise Sales Workshop for Founders with Peter Kazanjy on Wednesday, November 11, from 5:00 PM to 8:30 PM PST at 810 Hamilton St in Redwood City, with tickets starting at $50 according to the event page. You can browse everything else on the Tribe Events calendar on Luma.
Find your people
If you are weighing build-versus-buy decisions for your own AI stack, the AI Trends Tribe is a place to compare notes with other founders and builders. Tribe is ad-free and built around real conversations, so the discussion stays focused on what people are actually shipping.
