When you're developing AI for real-world environments, one of the biggest bottlenecks is getting the right data, from the right context, at the right time. We looked to industries solving similar problems—autonomous vehicle companies outfitting entire cars with sensor rigs, Google’s Street View backpacks, and even Apple’s private gym where they collect training data for the Apple Watch. The takeaway was clear: if you want accurate, reliable AI, you need a purpose-built way to collect real-world signals. So, we built FieldKits.

FieldKits are modular, easy-to-deploy sensor packs designed to collect audio, motion, and image data in rugged, real-world conditions. Built around Raspberry Pi boards with a range of interchangeable sensors, each kit can be quickly mounted, powered, and deployed in the field—whether that’s inside a vehicle, on construction equipment, or in an off-grid location. They stream data wirelessly and support multi-sensor logging and live annotation, making them incredibly useful for rapid prototyping and early-stage data collection.

These kits filled a critical gap in our process: they let AI teams move fast without waiting on production hardware. We used them to collect high-quality, structured training data from real environments—helping us iterate on models, catch edge cases, and test performance under realistic conditions.

FieldKits became our own version of a mobile data collection lab. They allowed us to continuously refine our models with real-world feedback, and in doing so, helped accelerate our edge AI development far beyond what’s possible in a controlled lab setting.