Angelos Perivolaropoulos

elevenLabs

The Research and Applied AI Summit (RAAIS) is a community for entrepreneurs and researchers who accelerate the science and applications of AI technology. In the run up to our 10th annual event on June 12th 2026 in London, we’re running a series of speaker profiles to shed more light on what you can expect to learn on the day!

At RAAIS we have a focus on translating cutting edge technology and research into production-grade products for real-world problems.

Angelos is a research engineer at ElevenLabs, where he leads the research engineering teams for speech-to-text and text-to-speech. His work sits at the intersection of model quality, systems engineering, and product deployment - exactly the point where modern voice AI becomes a real-world engineering discipline rather than just a research demo. In speech, progress is not only about model performance in isolation. It is also about latency, reliability, evaluation, and how systems behave under production constraints.

From research engineering to production voice systems

At ElevenLabs, Angelos leads the teams behind speech-to-text and text-to-speech research engineering. That makes his work especially relevant at a moment when voice interfaces are becoming a more serious part of the AI stack. Building these systems well requires more than strong models. It requires engineering that can connect research progress to reliable deployment across real products and workflows.

This is one of the defining challenges in applied AI now. As models improve, the bottleneck often shifts from raw capability to system design: how quickly models respond, how consistently they perform, and how well they hold up in production environments. Speech systems make those trade-offs especially visible because users experience them directly.

Scribe v2 and real-time transcription

Angelos led the team and research behind the Scribe v2 and Scribe v2 Realtime transcription models, which achieved leading results on most third-party industry benchmarks.

That matters because transcription is not a solved infrastructure layer. It remains a technically demanding problem shaped by accuracy, speed, robustness, and deployment context. Real-time transcription raises the bar further by forcing systems to perform under tighter latency and systems constraints while still maintaining high quality. Work in this area speaks directly to the broader challenge of turning foundation model progress into production systems that people can actually use.

Why this matters now

Voice is becoming an increasingly important interface for AI, and that puts more weight on the engineering required to make speech systems dependable in practice. As companies move from model releases to production voice products, the hard problems increasingly sit in the interaction between research and infrastructure: evaluation, latency, reliability, and operating models under real constraints.

That is what makes Angelos’s work relevant beyond speech alone. It reflects a broader shift in AI, where the value of a system is determined not just by benchmark quality, but by whether it can be made fast, stable, and usable in the real world.

Angelos’s background

Prior to joining ElevenLabs, Angelos worked in cloud infrastructure engineering, which shaped his practical approach to building reliable systems. That background is useful context for his current role, where research engineering depends not only on model development, but on the ability to make advanced systems work consistently in production.