Root Signals, a startup based in Helsinki and Palo Alto specializing in generative AI (GenAI) measurement with LLM-as-a-judge techniques, has raised €2.5 million. The round was led by Angular Ventures, with participation from Business Finland. Root Signals will use the funding to accelerate their platform and model development, and sales and marketing capabilities. Root Signals helps businesses accelerate GenAI adoption by providing them with enterprise-grade tooling to comprehensively measure, control and monitor LLM applications.
To safely ship GenAI-powered business applications to production, AI engineers have begun to use guardrails to block unintended behaviors. This approach controls risks but provides little capability for maximizing value. Instead, many are realizing they need to scrutinize the AI outputs with an extensive list of measurements with automations that mimic a human reviewer.
Ultimately, a task this complex can only be automated by other AI models, a technique often called LLM-as-a-judge. It allows one to quantify, for example, the degree of hallucinations, relevance of replies, or regulatory compliance. However, current AI frameworks do not adequately address the complexity, costs and unreliability involved.
A scalable, long-term approach to this is what the company calls EvalOps. By making it easy to build and automate these complex measurements, Root Signals makes production-grade GenAI applications quantifiable, reliable, auditable, and reusable.
“GenAI has no built-in quality control. You cannot treat it as traditional software, but rather you need to think of it as an unreliable freelancer. You have to be pedantic in instructing it, and then check its work in seven different ways – and then check again tomorrow. We make this scalable with metrics that are understandable and easy to maintain in production. Most other power tools in this sector are overly low-level and complex, or they provide more black boxes that kick the reliability can down the road,” said Dr. Ari Heljakka, Founder and CEO of Root Signals, with a PhD in GenAI.
Faster models
The most eager adopters of Root Signals have been independent software vendors providing GenAI-powered vertical bots to their specific domains of expertise, AI teams of fast-moving incumbent industry players seeking to develop competitive advantage, and LLM software consultants. With Root Signals, companies can build comprehensive metrics quickly, making detailed model-to-model comparisons easy. This unlocks a principled way to replace large models like GPTs with smaller, faster on-premise models – crucial for enterprises in regulated industries.
“Our evaluations distill the best practices and insights of essentially over 50 papers of recent years,” commented Oguzhan Gencoglu, Head of AI at Root Signals. “While measuring AI behavior is one thing, our users constantly ask: ‘How can I or my customers trust your AI itself?’ So, unlike other players, we baked self-measurability into the core of our evaluation engine.”
“Root Signals’s approach of using AI to manage enterprise AI implementation makes intuitive sense,” added Gil Dibner of Angular Ventures. “Everyone knows 90% of enterprise GenAI projects are stalling. To succeed, enterprises will need to implement LLM-specific evaluation tooling, which is not easy to begin with. Doing this well enough for enterprise use cases means building a robust constellation of LLM judges, and few enterprises have sufficient know-how to do this. Fortunately for them, the Root Signals founders have been thinking about this problem for 20 years.”
