Trajectory raises $45M to let professionals train AI without coding

Three Stanford alums launch Trajectory after DeepMind and Apple exits

Ronak Malde, Michael Elabd, and Arjun Karanam met at Stanford and left jobs at Google DeepMind and Apple to start Trajectory about a year ago. Sequoia just led their $45 million Series A round, pushing the startup’s valuation to $300 million. Nvidia, Bessemer Venture Partners, and Bain Capital Ventures also joined the deal.

Malde was part of the Windsurf team that DeepMind acqui-hired; Elabd built models at DeepMind, and Karanam worked on generative foundation models at Apple. Their goal: remove the need for coding so lawyers, doctors, and financial analysts can train AI models themselves.

Real-time feedback loops replace manual AI retraining

Most AI models stagnate after deployment—they only improve when engineers manually retrain them. Trajectory changes that by building feedback loops directly into everyday apps. A business starts with an open-source model, connects it to its tools, and collects real-time feedback on every task. The platform then retrains the model automatically, so it gradually performs more like a junior employee learning on the job.

“We eliminate the need for expensive consultants or in-house research teams,” Malde said. “A lawyer reviewing contracts or a financial analyst generating reports can now tweak the model themselves.”

Customers train AI models in days, not months

Harvey, Rogo, Decagon, Mercor, and Nvidia’s developer teams are already using Trajectory. One customer trained eight complex agent-based models in just two weeks. Another, Nvidia, worked with Harvey to adapt the newly released Nemotron 3 Ultra model for legal workflows—just 24 hours after its launch.

Users start with an open-source model, connect it to their daily tools, and let the platform’s feedback loops refine it over time. A legal-tech startup now has an AI that reviews contracts almost as accurately as a human lawyer, while a fintech firm trains assistants to mimic seasoned financial analysts.

A single dashboard replaces multiple AI training tools

Trajectory’s dashboard packs four steps—sandbox testing, evaluation, training, and deployment—into one screen. The first feature, Continual Grading, watches how a domain expert scores a handful of outputs, then applies those rules to thousands of unseen cases. The second, Group Size 1, keeps the model learning from every real-world interaction after deployment.

Instead of stitching together multiple tools, users integrate the platform with only five lines of code. The company says this cuts costs and friction, letting small firms compete with giants that can afford dedicated AI teams.

Competition grows as Applied Compute hits $3.25 billion valuation

Trajectory competes with Thinking Machines Lab, Applied Compute, and Prime Intellect. Last month, Applied Compute reached a $3.25 billion valuation, signaling how fast the market for post-training tools is growing.

The founders say their edge comes from hiring AI researchers and product designers from OpenAI, Stripe, and Netflix—people who understand both advanced AI and the pain points of everyday users. With the new funding, they plan to expand the platform and deepen integrations with popular open-source models.

What this means

Lazyfounder analysis — our interpretation, not reported fact.

Trajectory’s funding and valuation show investors are betting on tools that let professionals—without coding skills—refine AI models on the fly. If the platform delivers, it could move AI customization from research labs directly into law firms, clinics, and trading floors.

Key takeaways

• Trajectory’s platform lets lawyers or analysts improve AI models without hiring data scientists.
• One customer trained eight complex models in two weeks; another adapted Nvidia’s Nemotron 3 Ultra for legal work in 24 hours.
• The dashboard consolidates four steps into one interface, requiring only five lines of code.
• At a $300 million valuation, Trajectory enters a market where Applied Compute is already worth $3.25 billion.

Related on Lazyfounder

• Nvidia
• More ai news