Multimodal AI Model Optimization Research Engineer

Tavus•San Francisco, CA

6d•Hybrid

About The Position

At Tavus, we're building the human layer of AI. Our mission is to make human-AI interaction as natural as face-to-face interaction, enabling the human touch where it has been previously unscalable. We achieve this through pioneering research in multimodal AI for modeling human-to-human communication (language, audio, and video), as well as generating audio-visual avatar behavior. Our models power everything from text-to-video AI avatars to real-time conversational video experiences across industries like healthcare, recruiting, sales, and education. By enabling AI to see, hear, and communicate with human-like authenticity, we're creating the foundation for the next generation of AI employees, assistants, and companions. We are a Series B company backed by top investors, including Sequoia, Y Combinator, and Scale VC. Join us in driving the future of human-AI interaction. We’re looking for an experienced Research Scientist/Engineer with a focus on model optimization to join our core AI team. Our ideal partner-in-crime thrives in startup environments, is comfortable prioritizing independently, and is willing to take calculated risks. We’re moving fast and looking for people who can help pave the path.

Requirements

Strong experience in deep learning using PyTorch
Hands-on experience with model optimization and compression, including knowledge distillation, pruning/sparsification, quantization, and mixed precision
Understanding of efficient architectures such as low-rank adapters
Strong understanding of inference performance and GPU/accelerator fundamentals
Strong Python coding skills and reliable research engineering practices
Experience working with large models and datasets in cloud environments
Ability to read ML papers, reproduce results, and adapt ideas
Clear communication and collaboration skills

Nice To Haves

Optimization of diffusion models, video/audio generative models, or large language models
Experience with real-time or streaming systems (low-latency APIs, WebRTC, streaming TTS/video)
Familiarity with TensorRT, ONNX Runtime, TVM, Triton, or XLA
Experience writing custom Triton/CUDA kernels or low-level performance tuning
Experience with experiment tracking, benchmarking, and profiling at scale
Prior experience in research engineering or applied science roles

Responsibilities

Take cutting-edge research models and make them fast, efficient, and production-ready using sparsification, distillation, and quantization
Own the optimization lifecycle for key models: define metrics, run experiments, and benchmark trade-offs across latency, cost, and quality
Partner closely with researchers and engineers to turn new ideas into deployable systems