Genpio

Senior AI Engineer, Avatar & Voice Systems

Our avatar and voice cloning models are already trained, already good, and already run many live streams at once. You take them over: keep them fast and…

Engineering · Onsite, HCM, Vietnam · Full-time

About the role

We're hiring a Senior AI Engineer, Avatar & Voice Systems to take ownership of the models already running Genpio in production: the talking-head, lip-sync, and voice cloning stack that lets an AI avatar go live on TikTok and sell for hours without a human in the seat.

Read that carefully, because it changes the job. The models are trained. They already produce broadcast-quality avatars, they already clone a voice from short reference material, and they already drive many live streams at the same time. We are not asking you to start from a blank repo and a research paper.

We're asking you to own what exists and make it better. That means holding the latency budget as load grows, driving down the GPU cost of every stream-hour, squeezing more quality out of the models we have, onboarding new identities and voices without regressions, and keeping the whole thing up while real customers are live on it. There is no separate platform team to hand the pager to. You make the calls, and you live with them in production.

What you'll do

Own the models in production

  • Take over our avatar, lip-sync, and voice cloning models and be the person accountable for how they behave live
  • Know the current stack cold: where the time goes, where the memory goes, where the quality breaks
  • Keep round-the-clock streams healthy, and catch quality regressions before a customer does

Make it faster and cheaper

  • Hold a hard end to end latency budget: audio in, lip-synced frames out, live, with no drift
  • Cut cost per stream-hour through quantization, distillation, better batching, and killing kernel and CPU overhead
  • Push throughput and GPU memory further so the same fleet carries more concurrent streams

Improve the quality

  • Fine-tune and adapt what we have: better mouth detail, better identity retention, better prosody, without retraining the world
  • Onboard new avatars and new cloned voices from short reference material, fast and repeatably
  • Build the eval and regression loops that prove a checkpoint is actually better, then ship it behind a canary

Scale the serving stack

  • Grow multi-stream serving: continuous batching, multi-tenancy, admission control, graceful degradation
  • Design what happens when a GPU dies, a stream stalls, or demand spikes, and make it boring
  • Decide what's worth building, buying, or fine-tuning next, and defend it with numbers

What we're looking for

Must have

  • 5+ years in ML engineering, with deep PyTorch: you profile it, you read the trace, you don't just call .fit()
  • Real experience owning a model in production, not just training one: serving, latency, cost, on call
  • Inference optimization in anger: quantization, distillation, batching, CUDA graphs, TensorRT or similar
  • Comfort improving and adapting someone else's models and codebase rather than rewriting them
  • Able to make architecture calls alone and explain them clearly in English

Nice to have

  • Talking-head, lip-sync, TTS, voice cloning, or real-time avatar work specifically
  • CUDA or Triton kernel authoring, and comfort reading a profiler trace
  • Large-scale GPU serving (Triton Inference Server, vLLM, Ray Serve, or your own)
  • Experience running many concurrent generative streams on a fixed GPU fleet

Tools you'll use

PyTorch, CUDA, TensorRT, ONNX, Triton, Docker, Kubernetes, NVIDIA GPUs (A100/H100 class), Weights and Biases, Git/GitHub, Notion, Slack.

Interview process

  1. 30-min intro with an engineer
  2. Deep technical conversation: inference, serving, and where your last production system broke
  3. Practical exercise: profile and optimize a real inference path under a latency budget
  4. System design plus team fit: multi-stream avatar serving at scale, with the founders
  5. Offer

Compensation and benefits

  • Senior-level base plus meaningful equity in a fast-growing AI company
  • Onsite in Ho Chi Minh City with a modern, well-equipped office
  • Serious GPU budget and top-tier hardware, because the work needs it
  • Health coverage, team offsite, and a personal learning and conference budget

How to apply

Apply on this page. If you prefer email, send your CV to [email protected] or [email protected] with the subject [AI Engineer] Your full name, and tell us in a few lines about the work you are proudest of, with links where you can.

Why Genpio

Most AI engineering jobs hand you a research problem and a long road to anything real. This one hands you working models, live streams, and paying customers on day one. The interesting part is not getting it to work, it already works. The interesting part is making it fast, cheap, and beautiful at a scale almost nobody has run this kind of model at.

Apply, and see every open role: https://genpio.com/careers