Applied ML × Optical Imaging

Raj Singh-Moon, PhD

Director of Data Science · Modulim

Shipping ML in FDA-cleared medical devices. Research → production — from optical physics to models that make it onto the device.

Based
Costa Mesa, CA
Stack
Python · PyTorch · CUDA · C#
Background
Medical imaging · Columbia EE PhD
Papers
10+ peer-reviewed
01 · About

Engineer, scientist, developer — one person, all three.

Hybrid scientist-engineer. I build multimodal sensing and ML systems end-to-end — sensor integration and calibration, data processing, model development, failure analysis, and edge deployment — for clinician-facing capabilities in physiological assessment and triage.

PhD, Columbia EE (2019). Dissertation on biomedical optics and computational imaging — light-transport simulation and inversion validated against bench-top phantoms. MSc, Columbia EE (2016). BSc, NYU EE (2012). Ten-plus peer-reviewed papers across optics, clinical imaging, and ML.

Full-stack across the stack. Optical physics and CUDA Monte-Carlo simulation, CV/ML pipelines running on CM4/CM5, Jetson, and Luxonis OAK, production C# for device control, and the team and roadmap behind it all — I operate comfortably at each layer.

Operating ranges

  • Biomedical sensingEnd-to-end
  • Optical physics · CUDAExpert
  • ML / deep learningExpert
  • Edge / embeddedCM4 · Jetson · OAK
  • FDA-cleared workflowsShipped
  • Team leadershipDirector · 1 → N
02 · Selected work

Models that ship — measured by what they moved.

A selection of production and research work across imaging, ML, and team building. Impact numbers are program-level outcomes, not offline benchmarks.

01 · Case

Physics-informed CV & neural inverse solvers

End-to-end deep-learning models trained against the light-transport forward model — spectral reflectance to absorption and scattering coefficients in sub-ms, with physically consistent predictions on-device.

~60× speed-up~30% accuracy gain
Physics-informedPyTorchCUDAInverse problems
02 · Case

Object detection & semantic segmentation

Detection and segmentation models that localize regions of interest and classify tissue in multispectral captures — gating downstream inversion so oxygenation and tissue-structure outputs land on the right pixels.

FDA-clearedClinician-facing
DetectionSegmentationMultispectral
03 · Case
acqfiltersegfeatclsrawpredE2E BYPASS

Real-time edge deployment

End-to-end inference on RPi CM4/CM5, Jetson Nano, and Luxonis OAK — power, memory, and performance optimized so the full oxygenation pipeline runs on-device.

On-deviceCM4/CM5 · Jetson · OAK
Edge MLQuantizationEmbedded
04 · Case

Data-science function from scratch

Built and lead a cross-functional team spanning ML research, embedded integration, and product software — research to shipped device feature, in-house.

Director1 → N
LeadershipHiringMLOps
05 · Case
min

Applied GenAI for clinical workflows

Foundation-model integration with LoRA adapters and supervised fine-tuning — powering capture guidance, automated triage outputs, and reporting and review tooling.

LoRA · SFTWorkflow automation
LLMsFine-tuningClinical UX
06 · Case
epidermisdermissub-Q

GPU Monte-Carlo light transport

CUDA photon-transport simulator for layered tissue — the ground truth behind every optical-property inversion, from PhD bench work through production inference.

CUDA · C++Radiative transfer
CUDAC++Optics
07 · Case
live · 60 fps

Multi-sensor embedded integration

Camera sensor boards, custom stereo depth modules, ADC temperature sensing, DAC LED current control — photometric calibration and acquisition synchronization across the full optical chain.

Stereo visionPhotometric calibration
EmbeddedSensorsCalibration
03 · Experience

Fifteen years, one direction.

Lead a cross-functional effort to prototype and ship clinician-facing capabilities for a mobile vascular imaging device — oxygenation assessment and burn triage, from sensor to model to on-device inference.

  • Built end-to-end physics-informed computer-vision and neural inverse-solver models for the mobile imaging device — constraining training with the light-transport forward model so oxygenation and tissue-structure predictions stay physically consistent.
  • Shipped CV/ML pipelines for oxygenation inference and tissue-structure analysis, running on-device from capture to clinician-ready output.
  • Integrated multi-sensor hardware — camera sensor boards, custom stereo depth modules, ADC temperature sensing, DAC LED current control — on RPi CM4/CM5, Jetson Nano, and Luxonis OAK.
  • Built photometric calibration and acquisition synchronization across the full optical chain, with real-time edge optimization for power, memory, and performance.
  • Delivered clinical decision support and workflow automation — capture guidance, automated triage outputs, reporting and review tooling — plus the backend and frontend apps operating the device.
  • Applied GenAI: foundation-model integration with LoRA adapters and supervised fine-tuning for clinical workflow automation.
  • Own hiring, technical direction, and career development across ML, embedded, and product software.

Led production software and algorithm work for medical device control, acquisition, and inference — OOD, CI/CD, Agile/Jira — with authored system and software design docs.

  • Developed object-detection and semantic-segmentation models for the imaging pipeline — localizing regions of interest and segmenting tissue classes to gate the downstream optical-property inversion.
  • Delivered ~60× computational speed-up and ~30% accuracy gain via algorithmic redesign and GPU optimization.
  • Shipped production-grade C# for device control and data acquisition, with validation planning and root-cause investigations feeding the regulatory path.

Built image-quality assessment pipelines for an FDA-cleared imaging device, and translated Python/MATLAB research prototypes into production C# under TDD.

  • Authored the image-quality pipeline that became the cleared-device acceptance test, closing the loop between research and release.
  • Established the Python/MATLAB → production-C# translation pattern later teams standardized on.

GPU-accelerated real-time imaging analysis in the Structured Functional Imaging Lab, paired with experimental validation studies.

  • Built a GPU imaging-analysis interface for real-time reflectance and OCT data collection.
  • Ran experimental validation studies that grounded the computational pipeline in physical measurement.

PhD in Electrical Engineering — biomedical optics and computational imaging.

  • Built light-transport simulation and inversion machinery (Monte Carlo / radiative transfer) in CUDA, validated against bench-top tissue phantoms.
  • Designed and operated the optical and imaging hardware behind those studies — catheter-based probes and spectroscopic front-ends.
  • Co-authored peer-reviewed work on near-infrared tissue quantification and microcirculatory imaging.
06 · Contact

Open to senior-IC and Director-level roles in applied ML, biomedical sensing, and edge intelligence.

Happy to talk about ML against physical systems, shipping inference into regulated devices and onto the edge, and building cross-functional teams that take research all the way to clinician hands.

contact@rajinder.dev