Research Engineer

- Own evaluation pipelines — design, build, and automate offline and live evals that keep our speech and multimodal models honest in production.

- Harness the data — create tooling for safe, versioned, privacy-aware dataset curation and discovery.

- Ship models, not slide decks — partner with research and infra to prototype, train, and deploy state-of-the-art voice models that power Sesame’s real-time companion https://www.sesame.com/research/crossing_the_uncanny_valley_of_voice?utm_source=chatgpt.com experience.

- Squeeze silicon — scale training and inference for LLM-class workloads; chase latency, throughput, and cost until the graphs flatten.

- Wire up monitoring and live evals — surface quality regressions before users or PMs notice.

- Move at startup speed — take ideas from whiteboard to production in days, not quarters; leave a clean trail of tests and dashboards behind.

- Proven software engineer who loves ML; comfortable writing production code across the stack.

- Hands-on experience training or fine-tuning large language or other large-scale models with a variety of techniques.

- Evaluation expert — you’ve designed metrics and harnesses that actually predict user happiness.

- Deep knowledge of the ML lifecycle: dataset ops, training pipelines, eval frameworks, deployment, and monitoring.

- History of shipping complex projects to production—especially user-facing, online ML systems—despite shifting requirements and surprise roadblocks.

- High agency and the judgment to know when to sprint solo vs. pull in the squad.

- Track record of setting technical direction, driving consensus, and partnering smoothly with product, infra, and research.