Here's the honest version of the job.Most ML roles end when the model leaves the training environment. Someone else deploys it. Someone else finds out it's 40ms too slow on a phone. Someone else discovers the eval didn't match production.This role owns all of that.You'd own the models behind Moss: embeddings, rerankers, multilingual retrieval, and the intelligence layer in our Founding Agent. Dataset creation to training to evaluation to deployment to iteration. The full loop.The part that makes it hard: our models don't run on a GPU cluster behind an API. They run inside the agent runtime. In a browser. On an ARM device. On CPU. Sub-10ms is the budget, and the model has to fit inside it.So distillation and quantization aren't optimization work you do at the end. They're the job.What we're looking for:You've trained, fine-tuned, and shipped models to production. Not just notebooks.You understand representation learning and information retrieval deeply enough to know why a benchmark lies.You can navigate quality vs latency vs memory tradeoffs without someone handing you the answer.Bonus if you've touched Rust, profiled a system, or shipped anything to a mobile or browser target.What you'd get:Founding-level ownership and equity.Direct work with me, the runtime team, and the SDK team. No layers.Production failures as your dataset. Every miss in a real voice agent becomes a training signal you get to act on.SF or remote. We sponsor visas.By day 90, you'd own a meaningful part of the ML roadmap, not a ticket queue.If you want to build models and own what happens after they leave training, I'd like to talk.

ML Engineer (Founding Agent Models, Embeddings & Rerankers)

Emma

Responsibilities

Skills wanted:

Language(s) required:

Sri Malireddi

Sri Malireddi

About Moss Adams:

www.mossadams.com/

Admin access needed