ML Researcher

About Lightberry Lightberry builds the software layer for expressive, interactive robots. We combine voice AI, perception, and behavior systems so robots can understand people and respond naturally. We are a small, hands-on team based in San Francisco, working across software, hardware, design, and research to turn new capabilities into robots people can interact with today. We care about thoughtful craft, fast iteration, and shipping real systems into the world. Robots with Souls Yep, you read that right. Robots are here, but they ship with no software. They’re all mute, deaf, and blind out of the box. The only way to get the robot to do anything is to write code. This is insane. We want robots to behave like in the movies: proactive, caring, aware. The role The models that make a robot feel alive don’t exist yet — the perception, the timing, the sense of when to speak and when to wait. You’ll invent them. This is research with a heartbeat: your models ship onto robots that stand in real rooms with real people, days after they train. What you’ll do Research and train models for perception, behavior, and expression. Compress and deploy them to run on-robot, in real time. Design evals where the metric is “does this feel alive?” — then make it measurable. Work with animators and engineers to turn model outputs into character. Publish when it helps. Ship always. What’s needed We don’t care about degrees or citation counts. Show us what you’ve built and what you’ve figured out. We do need: Research. You’ve trained models that worked — and you can explain why they worked. Engineering. Your research runs outside a notebook: you write real code and ship it. Real time. You care about latency and footprint — your models run on a robot, not a cluster. Fun. You read papers for fun and reimplement the good ones. Bonus points Multimodal. You’ve worked with audio, vision, or motion models — ideally more than one. Robots. You’ve deployed models on hardware. 你会说中文吗? How to apply Tell us about something you’ve worked on before that you’re proud of Papers, models, repos — send what you’ve made, not just what you’ve read Show, don’t tell Keep it short One last thing… DO NOT apply if… You can’t stand chaos You’d work remotely every day if you could You’re not a people-person You’re pessimistic about tech, AI, and the future PLEASE apply if… You want to do your life’s work You care about developing and expressing your craft You love intense, fast-paced environments You like getting your hands dirty, no work is beneath you Life’s short, why work on B2B AI SAAS infra slop when you could work on speaking robots?