Location: Atlanta, GA - In-Person
OXOS empowers every provider with the capability, clarity, and confidence to make accurate decisions at the point of care. We are developing innovative radiographic imaging devices that push the boundaries of previous solutions by enhancing image quality, reducing radiation exposure, improving ease of use, and creating solutions to deliver care beyond traditional scenarios. We enable anyone, anywhere, to access radiologic diagnostics at the point of need, expanding availability and changing how healthcare is delivered.
OXOS is growing rapidly and is looking for a Machine Learning Engineer to join our Device Software R&D team and solve the deep technical challenges that make our imaging systems accurate and reliable — from correcting electromagnetic tracking distortion in real clinical environments to guiding operators toward correct anatomical positioning.
Responsibilities
As the Machine Learning Engineer, you will own the machine learning models behind two of our core device capabilities: the Electromagnetic (EM) Tracking system that locates our handheld X-ray emitter relative to the image cassette, and the View Confirmation software that guides operators toward correct anatomical positioning before an X-ray is taken. Your primary focus is advancing EM tracking distortion correction to a high-precision target of ±10 mm and ±1° despite real-world metal interference, and taking View Confirmation from a research prototype to a validated, production-ready tool running on our device hardware. As our research advances, you'll also have the opportunity to explore Noise-to-Noise (N2N) image denoising as a future direction. This is an R&D role centered on deep, 0-to-1 engineering problems on real hardware — success here means thriving on unsolved generalization challenges rather than implementing established solutions. Your responsibilities will include, but are not limited to:
Advance EM Tracking Generalization
- Pivot the EM tracking methodology toward broad environmental generalization, ensuring robust distortion correction across unseen metal configurations and clinical environments rather than only known interference sources.
- Establish a standardized "environmental stress test" evaluation across unseen metal configurations, targeting under 10 mm / ±1° median error.
- Design and implement a high-velocity data collection protocol that efficiently spans the full orientation problem space, cutting environment-specific data collection time to roughly 60 minutes per site.
- Port the encoder-decoder attention architecture into the production pipeline and validate its ability to generalize across multi-environment training sets.
Own View Confirmation, from Prototype to Production
- Transition View Confirmation from a developer-laptop research environment to an integrated solution running directly on the device Jetson, optimized for real-time performance.
- Establish baseline detection and feedback performance for novice-user guidance across core hand/wrist views, defining clear success metrics.
- Expand guidance coverage to ankle and shoulder anatomy while maintaining detection and guidance accuracy parity with the hand/wrist baseline.
- Ensure model backends and guidance algorithms support the full breadth of anatomies on the product roadmap, delivering documented validation reports to support production handoff.
Investigate Next-Generation Image Denoising
- Evaluate whether modern learned denoising approaches can meaningfully improve fluoroscopy image quality at low radiation doses beyond our current 3x3 Gaussian filter and frame-averaging baseline.
- Build a high-integrity fluoroscopy dataset from static scenes to enable training and validation of candidate denoising models.
- Train and evaluate candidate architectures (self-supervised and supervised) against current baselines, screening for hallucinated structural artifacts.
- Deliver a technical recommendation report, including a formal safety analysis, on whether a learned denoiser should replace the current baseline.
Standardize ML Engineering Practices
- Own the end-to-end ML engineering process, extending reproducibility, data lineage, and validation standards across all active ML projects — including EM tracking, View Confirmation, and denoising.
- Establish a unified experiment tracking and data lineage standard so every trained model is traceable to its dataset version and training configuration.
- Build a distribution shift detection mechanism that flags live inputs deviating from the training distribution, serving as an early-warning system for model performance degradation.
- Implement a standardized packaging and validation protocol for ONNX exports that enforces unit-consistency checks and verifies models against training-time invariants.
Required Skills and Qualifications
- Bachelor's Degree or higher in Computer Science, Electrical Engineering, Robotics, Applied Mathematics, or a related field.
- Expert-level Python and PyTorch skills, with experience building and training deep models end-to-end
- Advanced experience with scikit-learn, XGBoost, and LightGBM applied to structured, sensor-driven pipelines.
- Advanced NumPy/pandas skills for processing wide tabular time-series data across multiple sensor streams (200+ columns, 4+ sensors).
- Expert application of spatial and grouped train/test splitting to ensure valid extrapolation testing and prevent data leakage.
- Advanced comfort with Docker, conda, git, and CI pipelines (linting, type checking, automated testing).
- Collaborates effectively with device software, hardware, and clinical stakeholders to move R&D into production.
Preferred Qualifications
- Experience with sensor fusion, particularly integrating IMU data into multi-sensor pose estimates.
- Familiarity with ONNX export and runtime evaluation (e.g., Triton, TensorRT).
- Experience instrumenting live models for distribution shift detection and telemetry.
- Background in image denoising (e.g., Noise-to-Noise or other self-supervised approaches).
- Experience in regulated, safety-critical, or medical device software environments.
Benefits and Perks
- Health, Dental, and Vision Insurance
- Competitive pay and equity in the company
- 401(k)
- The opportunity to work with an innovative, early-stage company that is changing medical imaging as we know it
- Endless opportunities for growth and development in a rapidly growing medical company