Research Communications Lead, Physical AI

About us We're solving AI's hardest problem: alignment. Encord is the data layer for Physical AI and enterprise, serving as core infrastructure for 300+ of the world's leading AI teams, including Woven by Toyota, Hitachi, Skydio, AbbVie and Dyna, Walden and Agility Robotics. From autonomous vehicles and robotics to sports and healthcare, we're solving AI's hardest problem: turning complex, real-world data into production-ready intelligent systems. We're a team of founders, scientists, quants, researchers, and former Olympians, who’ve helped build some of the most successful companies in Silicon Valley, and alumni of MIT, Stanford, UC Berkeley, Oxford, Cambridge, and Imperial. We’ve raised $110M from legendary technology investors, early backers of companies like Databricks, ElevenLabs and Vercel, and we’re growing at an extraordinary pace. We're looking for exceptional people who want to help build a category-defining company, operating with high urgency and taking real ownership to help shape the future of AI. The role We recently released Encord Data Labs : our research arm through which we'll be sharing some of the lessons learned from our internal research on how data actually moves model performance in the physical world. We're looking for our first Research Communications Lead, to build and drive our Data Labs research and publishing motion. You won't just be on the author list, you'll be identifying interesting problems and topics to explore, and working with the rest of our ML team to shape their results into research that helps progress our field. This is a front-row seat to some of the most important technical challenges in physical AI, with a direct line to our research team and co-founders, as well as chance to explore the end-to-end journey of building autonomous systems and getting them to work in the real world. A taste of what Data Labs has worked on and published in its first months: • EMG-1 : we released the largest open dataset of wrist EMG synced with egocentric video: 307 hours of real work captured on head and wrist cameras, plus a 10.4-hour lab set with optical mocap, over 30x larger than the next public dataset of its kind • EMG + vision fusion : by adding muscle signals to camera-based hand tracking, we were able to reduce median fingertip error by ~25%, and ~35% when hands were briefly occluded • Encord-REACH : we built a video-native 3D hand pose model for egocentric work scenes, now in production with robotics teams; it cuts wrist tracking error by 80% and joint error by 65%, is 30–70% more accurate on industrial gloves, and almost never swaps in a bystander's hand • Synthetic data for ANYbotics : using NVIDIA Cosmos Data Factory to turn 410 reference images of industrial sites into 1,500+ short clips, we lifted ANYbotics' inspection anomaly recall by 24%, for ~$55 of generation What you'll do • Own the Data Labs research ideation: identify interesting and important topics worth dedicating resources to, collaborate with the broader ML team on the research process, and articulate our findings in a clear and compelling way • Partner with customers' research teams on joint research and case studies relating to their data pipelines • Work hand in hand with our broader ML team to scope, structure and draft research on topics ranging from embodied data collection approaches, to data curation, VLAs and world models • Build Data Labs publishing machine: a calendar, a review process with research, and templates for posts, benchmark launches, dataset releases and paper companions • Distribute every piece where the broader field can find it: X, LinkedIn, Hugging Face, arXiv companions, newsletters, podcasts and conferences such as CoRL, IROS and NeurIPS • Measure what lands, from citations and researc