# Senior Machine Learning Engineer, Voice Agents - EMEA Remote at Hugging Face

> Own the architecture and roadmap of the open-source speech-to-speech library and the hf-voice realtime voice-agent product.

- Company: Hugging Face (huggingface.co)
- Location: Paris, Île-de-France, France (remote)
- Places: Paris, Remote, France
- Posted: 2026-09-02
- Level: Senior
- Work mode: Remote
- Apply: https://apply.workable.com/huggingface/j/9E2A4C02C7/
- Page: https://ye.saranatour.dev/jobs/senior-machine-learning-engineer-voice-agents-emea-remote-hugging-face-kn769a3dgfvx5bzgsx3q8bgm0x8hf2qj

## The posting

Senior Machine Learning Engineer, Voice Agents - EMEA Remote
Hugging Face · Paris, Île-de-France, France

At Hugging Face, we're on a journey to democratize good AI. We are building the fastest growing platform for AI builders with over 11 million users who collectively shared over 3M+ models, 1M+ datasets & 1.47M+ apps. Our open-source libraries have more than 600k+ stars on Github.

**About the Role**

We are building the open voice-agent stack for Hugging Face, and we are looking for a senior engineer to own a large part of it.Two things sit at the centre of this role. The first is [speech-to-speech](https://github.com/huggingface/speech-to-speech), our open-source library for realtime voice agents. The second is hf-voice, a new product that will let any developer build and deploy voice agents with their Hugging Face account.

The library already powers the Reachy Mini fleet and there is a [public demo](https://huggingface.co/spaces/smolagents/hf-realtime-voice) running on Spaces, so you won't start from a blank page. But almost everything about how this becomes a product developers rely on is still open, and you will have a direct say in it.

**Your missions:**

**- Own the Open-Source Library:**

- Take architectural ownership of large parts of speech-to-speech: pipeline design, latency budget, and the reliability of the realtime loop.
- Integrate new ASR, TTS and end-to-end speech models as they land, and keep the abstractions clean while the model landscape keeps moving.
- Review community PRs, triage issues, cut releases, and grow the group of contributors around the project.

**- Ship hf-voice:**

- Design the developer API and the streaming protocol: session lifecycle, transport (WebSockets/WebRTC), authentication, error semantics, versioning.
- Build the serving side: realtime inference on GPU, concurrency, autoscaling, observability, and cost per session.
- Work with the Hub and inference teams so that a working voice agent is easy to integrate into products and demos.
- Take the product from demo to production: load testing, SLOs, graceful degradation when a model or a network path misbehaves.

**- Work in the Open:**

- Write the docs, examples and templates that get a developer from zero to a running agent in minutes.
- Support the deployments already relying on the stack, starting with the Reachy Mini fleet.
- Talk about the work publicly if you enjoy it: blog posts, demos, conference talks. We cover the travel and the prep time.

**Requirements**

**What we're looking for**

- Senior engineer, able to own a substantial part of an architecture and drive it forward autonomously.
- Experience building developer-facing infrastructure at an AI or developer-tools company: inference APIs, agent infrastructure, or something comparable.
- Substantial open-source contributions to a Python library. Comfortable with async Python and distributed systems, including their failure modes.
- You have shipped something realtime: streaming, WebSockets or WebRTC, audio or video pipelines, live inference.
- Practical experience with LLMs or multimodal models in production. Clear written communication and a habit of collaborating async and in public.
- Motivated by voice and conversational AI.

**Bonus points if you have**

- Contributions to a voice-agent framework such as speech-to-speech, pipecat, LiveKit Agents, Vocode or TEN.
- Contributions to llama.cpp or another low-level inference runtime.
- Hands-on work with ASR, TTS or end-to-end