Remy AI

Computer vision for smarter & safer warehouses

Documenting human dexterity

Warehousing is a $500B industry in the US, and 90% of its operators already have security cameras installed. Yet fewer than 10% use them for anything beyond security. Meanwhile decision-makers sit in their offices all day frustrated with their inability to know what is actually happening on the ground, benchmarking operations across different sites, identifying bottlenecks, and holding managers on the ground accountable. Remy AI is the intelligence layer that plugs into your existing camera infrastructure to track the metrics that matter. We combine AI models, analytical dashboards, and hardware where required to deliver a turnkey solution that adapts to your infrastructure and specific needs. Recent projects - Trunk fill and dwell times: auto-measuring inbound and outbound trunk fill percentages and tracking how long trailers are sitting empty on a bay before loading or full before departure, helping tighten bay allocation and alerting managers when trailers have been sat for too long - Workforce productivity by zone: measuring hourly scan rates, headcount and hours logged in key areas, and wave completion times, giving managers an objective view of where teams are hitting targets and where performance is slipping - Advanced security detection: flagging segregation zone breaches, out-of-hours activity, unauthorized entries, and backlog build-up in designated areas, turning passive CCTV into an active loss-prevention layer Our privacy-first architecture guarantees anonymization, with end-to-end encryption in transit and at rest.

Robotics foundation models are bottlenecked by data. The internet is saturated with cooking tutorials and unboxing videos, but it has almost nothing on the tasks that actually matter for industry: wire harnessing, SMT rework, surgical suturing, mechanical assembly, food prep at scale. Lab teleoperation and handcrafted sims try to fill the gap, but they collapse to archetypal motions in sterile environments and miss the contact-rich, force-modulated messiness of how work actually gets done. We're solving this by going to the source: we pair head-mounted capture with wearable tactile sensors to record synchronized vision, proprioception, and contact forces from people doing real work on real job sites. Our privacy-first pipeline guarantees anonymization, with end-to-end encryption in transit and at rest. Our proprietary annotation stack delivers contact events, hand poses, and drift-corrected trajectories at a granularity that off-the-shelf tooling can't match.