Software Engineer, Applied AI & Multimodal Systems

About Robot Health™ Robot Health is building the world’s largest autonomous healthcare workforce, starting in the home. We develop and deploy physical AI to address the rapidly worsening staffing shortages for non-medical home care. We’re building out all the key pieces of an autonomous caregiver: emotional intelligence, memory and personality, and physical capability. Each is in some ways superhuman, but all are grounded deeply in elevating the humanity of professional caregivers, family members, and aging individuals themselves. The Role You’ll work at the intersection of AI, software, hardware, and service delivery, building the “robotic brain” that allows our robots to understand and respond to people in real environments. This is not a traditional robotics role. We are not primarily looking for classical roboticists from mechanical engineering, electrical engineering, controls, or signal processing backgrounds. We are also not looking for pure ML researchers or data scientists. The core work is applied software engineering: taking processed signals from audio, vision, memory, models, sensors, and user context, then turning them into useful, reliable, real-time robot behavior. You’ll be integrating across frontier models, robotic platforms, edge infrastructure, multimodal pipelines, and care workflows to build solutions that don’t exist yet. This isn’t a single-layer job. You’ll be stitching together the full composite: evaluating the right models, libraries, and tools; wiring up AI systems that run in real time; debugging messy behavior in the physical world; and shaping how everything comes together in a live caregiving environment. No prior robotics experience? No problem. We care much more about your ability to think from first principles, architect complex systems, investigate unfamiliar technical domains, and ship working software quickly. Your Responsibilities: Architect and optimize real-time voice and vision pipelines that drive robot-human engagement. Design and tune voice, visual, and physical interaction behaviors that allow the robot to interact naturally with multiple humans at the same time, not just like a 1:1 voice agent. Develop long-term and short-term memory persistence and retrieval logic to properly inform robot-human interactions. Turn processed signals — audio, vision, speaker identity, user state, robot state, environmental context — into useful application logic for the robot brain. Evaluate and integrate the right models, libraries, APIs, and technical stacks for speech, vision, reasoning, memory, and embodied interaction. Develop and deploy inference/application stacks on edge devices like NVIDIA Jetson. Build application-layer behaviors that make the robot feel context-aware, responsive, and useful in real home environments. Develop holistic applications on top of SOTA multimodal and robotic foundation models, humanoid or otherwise. Debug real-world multimodal behavior in messy environments with real users. Ship. Constantly. We deploy to real environments with real users. What We’re Looking For We care far more about how you think and build than what’s on your resume: AI-native workflow: You use AI coding tools like Claude Code, Codex, Cursor, or similar systems as a core part of how you work: not as an experiment, but as your default. You understand that the leverage comes from strong system design sense, taste, debugging ability, and knowing how to direct AI effectively. Strong system design instincts: You can take a fuzzy problem like “make body language-based communication in a multi-speaker conversation environment feel natural” and decompose it into concrete, buildable components. You think in architecture, state, interfaces, latency, failure modes, user behavior, and evaluation loops — not just code. Applied AI judgment: You do not need to train foundation models, but you should know how to work with them. You can evaluate models, libraries, APIs, and open-source tools; understand their tradeoffs; and integrate them into a real product system. First-principles investigation: You are good at figuring things out. When facing an unfamiliar problem, you can reason from the underlying goal, research the landscape, test options quickly, and form a practical point of view. Solid programming chops: You don’t need to be an expert in everything, but you need to be dangerous enough to build and ship full systems. Proficiency in Python is preferred but not required. Product-minded engineering: You care about whether the system actually works for users, not just whether the technical component is interesting. You can make pragmatic tradeoffs between elegance, reliability, latency, cost, and user experience. High agency: You can take a nebulous task and run with it. You don’t wait for specs. You form a point of view, move fast, and adjust when you learn something new. Fast learner, strong communicator: You pick up new domains quickly. You can explain your thinking clearly. You challenge ideas with substance, not ego. Genuine excitement about the mission: You think AI and robotics in healthcare is one of the most important things being built right now. You want to be in the room where it happens. Nice to Have Experience building LLM-powered applications with tool use, memory, agents, or multi-turn reasoning. Experience with real-time multi-modal pipelines, WebRTC, voice agents, and streaming inference. Experience with computer vision models or libraries such as YOLO, MediaPipe, OpenCV, VLMs, or multimodal models. Experience with inference/application stacks such as Jetson/JetPack, vLLM, Triton, TensorRT, ONNX Runtime, or similar systems. Experience building software for multimodal interaction, human-AI interaction, or context-aware AI products. Experience building software that has to work in messy real-world environments, not just controlled demos. Prior work in healthcare, elder care, home care, or other high-stakes, real-world deployment environments. Who This Role Is NOT For If you are primarily looking for a classical robotics role focused on mechanical design, electrical systems, controls, motion planning, navigation, or low-level signal processing, this probably is not the right fit. If you want to focus mainly on training ML models, data science workflows, offline analysis, or research papers, this is not the right environment. We are much closer to the application layer than foundational research. If you need detailed specs and well-defined tickets to be productive, this isn’t the right environment. We’re building things that have never been built; ambiguity is the default. If you want to work on a single, well-scoped problem in isolation, you’ll be frustrated here. We need people who can context-switch, collaborate intensely, and still drive things forward independently. If you are skeptical of AI-driven development, we’re not a fit. AI-native tooling is how we operate, full stop. If you optimize for process over progress, we are in the wrong stage for you. Interested? Reach out directly. No cover letter needed, just show us what you’ve built. Must be authorized to work in the U.S. without sponsorship.