Anna Vintsevska
Leonid VeresIn our previous piece about TeqMate AI, we explored the idea behind its first reasoning layer and how it helps AdOps teams analyze data faster and make more informed decisions. Since then, TeqMate AI has evolved into a unified system for AdOps teams.
Leonid Veres, an R&D Engineer at TeqBlaze, shared how TeqMate AI can automate routine checks and understand performance changes before they affect revenue. He also explained why AdOps tasks may require different levels of analysis and how the system adapts its processing depending on task complexity.
***
Anna: You've been involved in TeqMate AI from the engineering side for quite some time. How has your role evolved as the project grew from an AI assistant into a much broader system?
Leonid: TeqMate AI has become my main focus. It's not a one-person effort, though — there's a whole RnD team of engineers who build, test, and refine the system together with me. We work closely with clients to understand how TeqMate AI behaves in real revenue and traffic scenarios and what problems may arise. That input is then translated into changes to the system logic to improve its performance. So, if there's a new idea to test or a new workflow to build, I'm likely to look at that first.
The funny thing is that the more we develop, the more room for growth there is. One idea leads to another. One solved problem opens the door to new improvements. The more TeqMate AI grows, the more opportunities we find to make it smarter. You might think it is exhausting. But for me, every opportunity is an exciting challenge.
As an R&D Engineer, what part of TeqMate AI are you personally most proud of today?
Probably two things. The first one is how TeqMate AI maintains context. Most AI tools treat you as a stranger. We wanted users to feel that the system keeps the relevant operational context. This changes the interaction completely. The experience becomes less like asking a generic assistant and more like working with a system that understands you and your work priorities.
The second is the way TeqMate AI learns from its own decision-making process. As the solution became more complex, we realized that delivering a result wasn’t enough. We need to understand how the system achieved the result. When the outcome isn’t quite what we expected, knowing how it got there helps us improve the system. Over time, we can identify which diagnostic approaches work best for recurring AdOps scenarios and improve the system to better match those patterns.
When people hear "AI assistant," they usually imagine a simple system with limited capabilities. At what point did you realize TeqMate AI was becoming something much bigger than that?
The turning point came when we released the first MVP to a broader audience. Before that, we were using it internally. We were testing, refining, and challenging it inside our teams. Our goal was to validate how people communicate with smart assistants and what kind of assistance they are actually looking for.
Once our clients got access, we understood that real AdOps workloads are far more complex and require different types of analysis and the ability to connect information from multiple sources. That's when we started expanding TeqMate AI beyond the assistant model.
Today, TeqMate AI includes multiple analytical and operational capabilities. Was there a specific moment when you realized that a single AI model would not be enough to address operational team challenges?
For me, it was more of a gradual realization that AdOps work is extremely diverse. There is a clear difference between using a single AI approach and combining specialized capabilities. Asking one model to work out every problem feels like asking one employee to be an analyst, engineer, and accountant simultaneously. Technically possible, but it doesn’t match the complexity of AdOps workflows. A simple task doesn't need the same amount of processing as a complicated investigation — using a heavyweight model for a simple check just adds cost without adding value. So the model itself needed to be chosen based on the complexity and scope of the question: a lighter model for quick checks, a more capable one for deeper investigations. That gives us both a better result and a more cost-efficient service.
TeqMate AI now routes tasks dynamically. The system decides how deep to go, what data to collect, and which model should handle the response. Instead of being limited to predefined tasks, TeqMate AI puts together multiple components to handle more complex AdOps workflows. These include identifying top-performing DSPs, comparing today's performance with previous days, or detecting potential revenue drops. Users don’t have to manually analyze reports or compare charts — the system performs these comparisons directly and surfaces the key insights.
Have there been situations where TeqMate AI identified an issue that a human team would likely have missed?
Definitely. Not because people aren't good at their jobs, but because there is simply too much information. AdOps teams are great at spotting important signals. The problem is that they often have thousands of signals. Sometimes a team focuses on traffic quality while the actual issue comes from floor pricing, query volume, filtering logic, or some unexpected configuration conflict. Those aren't always obvious, especially when you are managing many accounts and workflows at once.
TeqMate AI doesn't get tired of checking all those variables. By applying task-specific logic for different types of work, it adapts to the nature of the task instead of forcing every problem through the same process.
If we have this conversation again in two years, what do you think TeqMate AI will be capable of that it cannot do today?
I think the biggest change will be moving from "here's a problem" to "here's a problem, here's why it happened, and here's what you should do next." Today, TeqMate AI is already helping users investigate causes faster. The next step is helping them notice issues before they even start looking for them. The direction we are heading in goes beyond issue detection. TeqMate will eventually be able to validate them, test hypotheses, and confirm what is actually happening before delivering conclusions.
I also expect proactive alerts to become a major part of the experience. Instead of waiting for users to ask questions, the system will automatically surface important insights. We're not fully there yet, but we're getting closer with every product release. But honestly, we'll probably have an even longer list of ideas than we have today. And that's the best part of working on a project like this.
***
As AdOps environments become more fragmented and signal-heavy, the focus shifts toward systems that can connect context across different actions and reduce the need to manually piece information together. TeqMate AI is gradually moving beyond a tool that responds to isolated requests and is becoming part of the daily operational flow, where teams already make decisions and react to changes.
TeqMate AI is already running in real environments, supporting teams in analyzing traffic, tracking revenue changes, and automating routine workflows. You can see these scenarios in practice in our webinar with the TeqBlaze CPO, Olha Zharuk, where we discuss AdOps use cases. In the session, we show how TeqMate AI reduces manual analysis and helps teams understand the root causes faster.
