Trajectory Secures $45M to Democratize AI Model Training for Nontechnical Users

Image: TechCrunch · Source

Trajectory has raised $45 million to enable businesses to continually improve AI models without technical expertise, using a platform that automates feedback-based retraining, aiming to make specialized AI accessible and affordable.

Trajectory, a one-year-old AI startup founded by former researchers from Google DeepMind and Apple, recently announced a $45 million Series A funding round that values the company at $300 million. The round was led by Sequoia Capital and included participation from Nvidia, Bessemer Venture Partners, and Bain Capital Ventures. Trajectory aims to transform how businesses adapt AI models for specialized tasks by simplifying the retraining process for nontechnical users.

Current AI foundation models, like those developed by OpenAI and Anthropic or open source alternatives, generally do not learn from their mistakes without further training. Companies often rely on costly, specialized engineers or internal research teams to fine-tune these models for specific fields such as law and finance. Trajectory’s platform automates continual learning—a method that allows AI models to improve over time by learning from ongoing feedback collected directly from users’ applications.

By connecting an existing open source AI model to Trajectory’s platform, businesses can gather performance data and enable the system to retrain the model iteratively. This approach helps experts in various domains, including lawyers and doctors, improve AI model performance on domain-specific tasks autonomously, reducing the need for highly specialized technical staff and accelerating deployment.

Clients like Harvey and Rogo have already utilized Trajectory’s technology to post-train AI models, enabling them to act more like experienced professionals by learning from specific tasks, such as reviewing standardized legal contracts or applying industry-specific valuation methods. Other customers include Decagon, Mercor, and Nvidia. Trajectory has recently made its post-training platform publicly accessible.

The company’s founders highlight the importance of human involvement in the training loop to maintain safety and control, distinguishing their work from autonomous recursive self-improvement approaches that could unpredictably alter AI behavior. The founding team combines deep technical expertise and a strong emphasis on user-friendly design to bridge the gap between sophisticated AI training techniques and practical business applications.

Trajectory competes with emerging startups such as Thinking Machines Lab, Tinker, Applied Compute, and Prime Intellect, all focusing on automating or simplifying AI model specialization and training. The company’s leadership, experience from top AI research labs, and its focus on accessibility and design are considered critical differentiators in this burgeoning market.

This development represents a significant step toward democratizing AI customization, allowing organizations without extensive technical resources to effectively harness and personalize AI capabilities for their unique operational needs, potentially shaping the future landscape of AI deployment across industries.