Particle.news

Trajectory Launches Self‑Serve Platform to Let Models Learn From User Corrections

The product promises to automate continual fine‑tuning using in‑product edits, cutting deployment time, lowering the need for specialist engineers, opening customization to nontechnical teams.

Overview

  • Trajectory launched its self‑serve continual learning platform on Monday, November 2, 2026, shifting from closed beta partnerships to public availability for qualifying businesses.
  • The company announced a reported $45 million Series A led by Sequoia that values Trajectory at roughly $300 million and follows earlier seed funding, bringing total disclosed investment to about $55–65 million.
  • A lightweight SDK captures user edits and corrections inside customer products and feeds those signals into regular post‑training cycles, and the product includes tools called continual grading and a GS1 algorithm to scale human feedback.
  • Early customers named by the company include Clay, Harvey, Decagon, Mercor and Rogo, and Trajectory lists Nvidia and major VCs as participants in its funding and partner list.
  • Enterprise buyers and industry watchers warn of privacy, consent, governance and feedback‑quality risks from training on production data, and Trajectory says it follows SOC 2 controls and retains human‑in‑the‑loop checks to reduce those risks.