AI Researcher (Multimodal Audio/Video Generation)

Tavus is a research lab pioneering human computing through AI Humans—a new interface enabling real-time, face-to-face conversations between people and machines. We're building neural avatars that see, hear, respond, and look real, combining emotional intelligence with 24/7 availability across languages. As a Series A company backed by Sequoia Capital and Y Combinator, we're shaping the future of human–AI interaction at scale.

What you'll do

• Lead research efforts on audio-visual generation for conversational avatars, including neural avatars and talking-head models.
• Design models that couple conversation flow with synchronized verbal and non-verbal signal generation.
• Drive innovation in diffusion models, long-video generation, and multimodal audio-visual modeling.
• Translate research into production by collaborating with Applied ML and engineering teams.
• Mentor researchers, set research directions, and publish impactful work in top-tier venues.

What Tavus is looking for

• PhD or equivalent research experience, plus 2–3+ years applying generative models at scale.
• Deep expertise in diffusion models and knowledge of efficiency techniques.
• Proven experience in multimodal generation spanning video, audio, and language.
• Track record of innovation in long-video or audio generation.
• Strong programming skills with fluency in PyTorch and GPU-optimized workflows.
• Published research in top-tier venues (CVPR, NeurIPS, BMVC, ICASSP, etc.).
• Experience leading research activities or mentoring teams.
• Nice-to-haves: 3D graphics, Gaussian splatting, large-scale training setups, or broad exposure to generative AI models.

Location: San Francisco (hybrid) or London preferred; remote within U.S. or Europe considered for exceptional candidates.