Trajectory raises $45M to make AI model training accessible to nontechnical experts

We are entering a phase of the AI cycle where the technical moat is shrinking. For the last two years, the conversation has been dominated by GPU clusters, context windows, and which model is winning the leaderboard this week. But if you talk to founders who are actually trying to deploy this stuff in a vertical industry, the problem is rarely the code. It is the data.

Specifically, the problem is getting the right people to label and refine that data. Trajectory, a startup that just closed a $45 million Series A, is betting that the next wave of value comes from letting subject matter experts steer the ship without having to learn Python.

The Value of the Domain Expert

In the early days of any tech shift, the builders are usually the engineers. They build tools for other engineers. But once the foundational layers are set—which is where we are with LLMs—the utility of the technology depends on how well it understands a specific niche. An engineer at a top-tier lab might know how to optimize a transformer block, but they don't know the nuances of maritime law, structural engineering, or specialized oncology.

Historically, training a model meant a messy back-and-forth between the expert and the data scientist. The expert explains the nuance, the scientist tries to translate that into a training set, something gets lost in translation, and the model comes out hallucinating in ways that are obvious to a professional but invisible to a coder. Trajectory is trying to remove the middleman.

By raising $45 million at a $300 million valuation just one year into their journey, they are signaling that the market believes "no-code" for AI training is the next major infrastructure play. If you can make model fine-tuning as simple as using a spreadsheet or a drag-and-drop interface, you unlock the institutional knowledge currently trapped in the heads of senior professionals who don't know what a learning rate is.

Why This Matters for Builders

If you are building a startup today, you need to look at this as a shift in where you spend your capital. In 2023, you spent it on expensive AI engineers. In 2025 and beyond, you might be spending it on the best subject matter experts you can find, using platforms like Trajectory to let them build the product themselves.

This is a democratizing force, but it’s also a competitive threat to the traditional "AI wrapper" companies. If every law firm can easily train its own proprietary model using its own internal partners as the trainers, why do they need a generic legal AI startup? Builders need to move further up the stack. Don't just provide the interface; provide the data pipeline that makes the expert's job easier.

• Lowering the barrier to entry: You no longer need a PhD-heavy team to create a specialized model.
• Reducing translation error: Direct input from experts means fewer iterations and more accurate outputs.
• Speed to market: Rapid prototyping becomes possible when the person with the idea can also be the person training the weights.

The Skeptical Take: Is This Just a UI Play?

As much as I like the vision, we have to stay grounded. Every time a new technology emerges, someone tries to sell a "no-code" version of it. Sometimes it works—think Webflow for web design—and sometimes it fails because the underlying technology is just too complex to simplify without breaking it.

Training a model isn't just about labeling data. It's about understanding bias, overfitting, and catastrophic forgetting. If you give a non-technical expert a powerful tool without the proper guardrails, they might build something that looks great in a demo but fails miserably in production. The $300 million valuation suggests investors think Trajectory has solved the guardrail problem, but for founders, the proof will be in the actual performance of the models these experts produce.

The real moat in AI isn't the model you use; it's the specific, hard-to-get knowledge you feed into it. If you make it easier to input that knowledge, the value shifts from the software to the person holding the expertise.

The Strategic Pivot

For those of us in the trenches building companies, this news should change your hiring roadmap. Instead of looking for your fifth machine learning engineer, maybe you should be looking for a retired specialist in your target industry. If tools like Trajectory deliver on their promise, the engineer becomes the plumber who keeps the pipes running, while the expert becomes the architect who actually designs the house.

We are seeing a move away from "General AI" toward "Deep Vertical AI." The winners won't be the ones with the most compute, but the ones who can most efficiently harvest the brains of the world's most experienced professionals. This $45 million raise is just the first shot in that shift.

What to Watch Next

Keep an eye on how the big players react. If OpenAI and Anthropic decide to build robust, no-code fine-tuning suites directly into their enterprise dashboards, startups like Trajectory will have to offer something significantly deeper than just a nice UI. They will need to offer better data management, better version control, and better ways to measure the "expert's" actual impact on the model.

For now, it’s a win for the builders. It means we can focus less on the syntax and more on the solution. If you've been sitting on a specialized dataset because you didn't have the technical team to exploit it, that excuse is officially expiring.

Read the original at TechCrunch Startups →