Articles / Trajectory Raises $45M for AI Model Post-Training Platform

Trajectory Raises $45M for AI Model Post-Training Platform

4 11 月, 2026 • 3 min read • AI-Infrastructurecontinual-learning

Trajectory Secures $45 Million in Series A Funding Led by Sequoia

In a major milestone for the AI infrastructure space, Trajectory, a startup specializing in post-training and continual learning for large language models (LLMs), has announced a $45 million Series A round, valuing the company at $300 million.

🔑 Key Investors: Sequoia Capital (lead), NVIDIA, Bessemer Venture Partners, and Bain Capital Ventures.

The funding underscores growing demand for tools that enable enterprises to adapt foundational models to domain-specific tasks—without requiring full fine-tuning expertise or massive internal AI teams.

Trajectory Post-Training Platform
Source: Trajectory


The Problem: Static Models vs. Dynamic Expertise

Today’s state-of-the-art LLMs—from OpenAI and Anthropic to open-weight alternatives—face a critical limitation: they cannot learn from real-world feedback after deployment. Without additional training, they remain static—even when making repeated errors in high-stakes domains like law or finance.

Traditional adaptation methods like fine-tuning often require:
– Dedicated AI research teams,
– Costly forward-deployed engineers (FDEs), or
– Deep technical expertise rarely available outside elite engineering orgs.

Trajectory aims to eliminate this barrier.


The Solution: Continual Learning as a Service

Trajectory’s platform enables organizations to:

✅ Start with any open-source LLM (e.g., Llama, Mixtral),
✅ Integrate the model into their application stack,
✅ Automatically collect performance signals (e.g., user corrections, task success metrics), and
✅ Use those signals to continuously retrain and refine the model—improving accuracy, domain fluency, and task-specific reliability over time.

💡 As co-founder Ronak Malde (ex-Google DeepMind) explained to TechCrunch:
“A foundation model is like a brilliant PhD who starts fresh every time. What creates economic value is someone who’s worked in the field for 30 years.”


Real-World Adoption & Differentiation

Several AI-native companies are already leveraging Trajectory’s platform:

  • Harvey and Rogo: Enhancing legal reasoning models via iterative contract review feedback.
  • Decagon, Mercor, and NVIDIA: Accelerating vertical-specific model optimization across finance and systems engineering.

Unlike fully autonomous recursive self-improvement (RSI), Trajectory’s approach preserves human-in-the-loop governance, ensuring safety, auditability, and alignment with organizational standards.


Competitive Landscape & Technical Edge

Trajectory operates in a rapidly intensifying segment. Key peers include:

Company Focus Notable Backing
Thinking Machines Lab (Tinker) Model tuning platform Founded by ex-OpenAI CTO Mira Murati
Applied Compute Enterprise AI infrastructure $3.25B valuation (Oct 2026)
Prime Intellect AI agent development & training Focused on bespoke agent orchestration

Yet Trajectory emphasizes its team-driven advantage: Founders from Stanford, Google DeepMind, Apple, and OpenAI—plus hires from Stripe and Netflix—bring deep ML systems, product design, and human-AI interaction expertise.

Crucially, its UI-first philosophy lowers the bar for non-technical subject-matter experts (e.g., lawyers, clinicians) to directly define performance goals—and translate them into measurable model improvements.


Public Launch & Strategic Vision

The company has officially opened access to its post-training platform, signaling readiness for broad enterprise adoption.

Malde affirms:

“Users define what ‘excellence’ means for their organization—and Trajectory turns that definition into tangible, evolving model capability.”

With continual learning poised to become a standard layer in the AI stack, Trajectory’s mission—to make expert-level model evolution accessible, safe, and scalable—has never been more timely.


Reference: TechCrunch — Trajectory Raises $45M to Make AI Model Training Accessible to Non-Technical Experts