Custom Model Building FAQ

Generic rPPG models are trained on standard webcam or smartphone camera data. Your hardware has a different image sensor, optics, noise profile, and operating environment. A custom model trained against your specific camera's characteristics extracts substantially more physiological signal, resulting in more reliable vital signs in real-world conditions on your device.

We build models for RGB, near-infrared (NIR), short-wave infrared (SWIR), thermal, and multi-spectral sensor arrays. Whether your device uses a standard CMOS imager, an automotive-grade NIR camera, a time-of-flight sensor, or a custom multi-spectral module, we can characterize it and train a model accordingly.

We design models to fit your compute budget, not the other way around. We have delivered models running on everything from high-performance automotive SoCs to resource-constrained IoT microcontrollers. During hardware characterization, we profile your target platform and architect the model to meet your specific latency, memory, and power constraints.

Most engagements run 6 to 12 weeks from hardware characterization to production delivery, depending on complexity. Projects involving standard RGB cameras with well-defined environments tend toward the shorter end. Novel sensor types, multi-spectral arrays, or particularly constrained compute targets may extend the timeline.

Yes. Every model we deliver is architected for fully on-device inference with zero cloud dependency. This is not optional — it is fundamental to our approach. Biometric data never leaves the device, satisfying automotive functional safety, healthcare data regulations, and enterprise privacy requirements.

Submit a custom build inquiry through our portal. Our engineering team will schedule a technical scoping call to understand your hardware, use case, and deployment requirements. From there, we provide a detailed project plan including timeline, deliverables, and pricing tailored to your specific build.