Trajectory Raises $45M to Put Continual AI Training in Experts’ Hands
Trajectory raises $45 million at a $300 million valuation to help nontechnical experts continually retrain open source AI models with human feedback safely.
Summary
On November 2, 2026, Trajectory announced a $45 million Series A at a $300 million valuation. Sequoia led the round, joined by Nvidia, Bessemer Venture Partners and Bain Capital Ventures. The 18-person startup was founded about a year ago by Stanford classmates Ronak Malde, Michael Elabd and Arjun Karanam. Malde joined Google DeepMind through its Windsurf acqui-hire, Elabd also worked at DeepMind, and Karanam worked on generative foundation models at Apple. Trajectory has also recruited from OpenAI, Stripe and Netflix.
OpenAI, Anthropic and open source models require further training to learn from mistakes. Trajectory connects open source models to applications, gathers performance feedback and retrains them continually, allowing lawyers, doctors and other nontechnical specialists to define successful results through an accessible interface. Harvey and Rogo use it for legal contract reviews and company valuations; Decagon, Mercor and Nvidia are also customers. The post-training platform is now publicly available. Trajectory keeps humans involved to reduce risks from autonomous recursive self-improvement. Competitors include Mira Murati’s Thinking Machines Lab and its Tinker platform, Applied Compute, valued at $3.25 billion last month, and Prime Intellect.
Positives
- $45 million in Series A funding gives the year-old, 18-person startup substantial capital for expansion.
- $300 million valuation signals strong investor confidence from Sequoia, Nvidia, Bessemer Venture Partners and Bain Capital Ventures.
- Harvey and Rogo already use Trajectory for specialized legal and financial tasks, while Decagon, Mercor and Nvidia are also customers.
- Lawyers, doctors and other specialists can guide model improvement without building internal AI research teams.
- Public availability opens Trajectory’s post-training platform to more businesses using open source models.
Risks & concerns
- OpenAI, Anthropic and open source models still require additional training to learn from mistakes.
- Traditional fine-tuning can require costly consultants known as forward-deployed engineers or dedicated internal research teams.
- Autonomous recursive self-improvement could create unintended or dangerous outcomes, making continued human oversight important.
- Thinking Machines Lab, Applied Compute and Prime Intellect create substantial competition in automated post-training.
- Applied Compute’s $3.25 billion valuation highlights the scale and funding strength of Trajectory’s competitive market.