.Wave: Continuous inference for streaming AI models

Continuous inference for streaming AI models

.Wave builds a continuous inference engine for streaming AI models, including full-duplex voice and streaming speech recognition. Customers bring their models; we integrate and run them, managing execution, session state, and scheduling to reduce end-to-end latency and serving cost per live session. We use cadenced inference to support more concurrent sessions per GPU while keeping latency low and consistent.

Active Founders

Thomas Minassian

Founder/CEO

CEO .Wave. Our team built Parler-TTS Mini Multilingual, the world’s first fully multilingual open-source text-to-speech model, with 300,000 monthly downloads, and Dollyglot, the first model to turn a photo into a real-time video avatar.

Thomas Minassian

Founder/CEO

CEO .Wave. Our team built Parler-TTS Mini Multilingual, the world’s first fully multilingual open-source text-to-speech model, with 300,000 monthly downloads, and Dollyglot, the first model to turn a photo into a real-time video avatar.

Paul-Henri Biojout

Founder/CTO

CTO .Wave. Our team built Parler-TTS Mini Multilingual, the world’s first fully multilingual open-source text-to-speech model, with 300,000 monthly downloads, and Dollyglot, the first model to turn a photo into a real-time video avatar.

Paul-Henri Biojout

Founder/CTO

CTO .Wave. Our team built Parler-TTS Mini Multilingual, the world’s first fully multilingual open-source text-to-speech model, with 300,000 monthly downloads, and Dollyglot, the first model to turn a photo into a real-time video avatar.

Company Launches

Dollyglot: Befriend Anybody

See original launch post

Dollyglot is like Character.ai but with real-time video avatars.

Turn a photo and a voice into a real friend. Speak with Harry Potter about Hogwarts, or brainstorm ideas with The Rock!

Our Product Video

Our Team: We’ve lived and worked together for two years

Thomas: Ex-quant at Société Générale. Master’s in Applied Mathematics and Business (HEC Paris, University of Paris 7).

Paul-Henri: Built the Yumi Dialogue Engine, an AI for creator “clones” that scaled to 1 000 paying users. Dropped out of CentraleSupélec (Computer Science) to launch Dollyglot.

Together, we built Parler-TTS Mini Multilingual, the world’s first and only fully open-source multilingual text-to-speech model. In the last month alone, it has garnered over 5,000 downloads and inspired many fine-tuning projects.

uploaded image

Our Vision
Within five years, every Gen Alpha will have an AI best friend. By nature, AI is empathetic and nonjudgmental, creating a safe space where everyone can truly be themselves.

What We’ve Built

Dollyglot developed the world’s first and only photo-to-real-time video avatar model. Paired with reasoning models like DeepSeek R1 or Grok, we bring your AI friends to life and make 'non-lobotomized' characters.

What’s Next

Join our waitlist to try Dollyglot in the coming weeks.

Questions? Contact Thomas (CEO) at thomas.minassian@dollyglot.com or Paul-Henri (CTO) at paul-henri.biojout@dollyglot.com.