[ML Engineer Application ]
WATER is a Biophysical Intelligence research company. Our mission is to teach matter to respond to life: we study how the human body and the physical world respond to each other, and we build machines that sense a body, model its state, and respond in real time. CAMA, an adaptive bed, and FLOW, an adaptive chair, are the first surfaces this research runs on. The work spans three research directions: machine learning and mathematics, bio research, and materials and electronics. The long-range outcomes we care about are human longevity and human experience.
You own the path from trained weights to deployed inference on real hardware. You'll work on quantisation and optimisation, the evaluation harness, latency and memory budgets, and monitoring of deployed models.
In This Role You Will
- -Take research models to production: quantisation, pruning, distillation, runtime selection, latency and memory budgeting on the target.
- -Own on-device inference alongside firmware — and the telemetry that comes back from it.
- -Build the evaluation harness. Make it fast, automatic and impossible to fool.
- -Profile and optimise. Know which layer is actually costing you the deadline.
- -Monitor deployed models for drift, degradation and silent failure, before a user notices.
- -Work with researchers early enough to tell them when an architecture will not fit the target.
What We Hope You'll Bring
- -You have put a model into production and kept it working.
- -2+ years in ML engineering or applied ML.
- -Strong Python and PyTorch; comfortable in C++ when the runtime demands it.
- -Real experience with deployment tooling — {ONNX, TensorRT, TFLite, ExecuTorch} or equivalent.
- -You have opinions about reproducibility and they are load-bearing.
- -Nice to have: edge inference on ARM-class hardware, streaming or time-series inference, embedded systems literacy.