Senior Computer Vision & Machine Learning Engineer

illumicell AI; United States; Remote; USD 140,000 - 190,000; Full time

Benefits offered by this job

Flexible paid time off

Job summary

illumicell AI is building a digital holographic microscope for automated semen analysis. Our instrument captures live video, reconstructs phase imagery in the cloud, and produces clinically relevant measurements used by fertility labs to guide treatment decisions.

We seek a Senior Computer Vision and Machine Learning Engineer to tackle tough imaging problems and deliver tested, production‑quality software. The role combines deep image processing work with software breadth across device

Qualifications

• Demonstrated experience building computer‑vision or image‑processing systems in production/clinical environments.

Responsibilities

• Take responsibility for CV capabilities from problem definition to tested product component.
• Develop, train, and deploy models for detection, tracking, motility, and image‑quality assessment using PyTorch.
• Design evaluation systems with disciplined dataset splits, versioned artifacts, and clinically relevant metrics.
• Define annotation requirements and collaborate with clinicians to build training/testing data.
• Carry models through GPU profiling, batch inference optimization, and deployment.
• Contribute to device acquisition software and cloud data handling when prioritized.
• Establish repeatable training workflows and model provenance.
• Review experiments and code with ML colleagues and provide mentorship.

Skills

Python

PyTorch

English communication

Software testing

Education

MS/PhD in CS/EE/BME

Tools

OpenCV

NumPy

SciPy

ONNX

TensorRT

CUDA

Job description

Who We Are

illumicell AI is a medical-technology startup building AI-powered, label-free cell analysis to bring fast, objective diagnostics to the point of care. Based at the Pagliuca Harvard Life Lab, Harvard's startup incubator in Allston, we work at the intersection of optics, machine learning, and clinical medicine.Starting with male fertility, we're developing a new standard for how living cells are measured, understood, and acted upon. We have a functional prototype device, a digital holographic scanner paired with cloud-based AI analysis that is already producing live-cell measurements, and we are rapidly accelerating toward FDA submission. We're backed by Accel and Tim Draper, are Techstars alumni, and are building the team that will carry the product from working prototype to cleared clinical device.

About the Role

illumicell AI is building a digital holographic microscope for automated semen analysis. Our instrument captures live-cell video, reconstructs phase imagery in the cloud, and produces clinically relevant measurements (sperm concentration, motility, and eventually morphology) that fertility labs use to make treatment decisions.We are looking for a Senior Computer Vision and Machine Learning Engineer to take technical responsibility for the path from difficult imaging problems to tested, production‑quality software. The problems are real and hard: dense fields of small moving cells, overlapping objects, reconstruction artifacts, limited ground truth, and the need to separate measurement error from sample and image‑quality failures. The work covers detection, multi‑object tracking, motion analysis, image quality, and evaluation design.The role requires depth in image processing and computer vision, combined with enough software breadth to contribute to device acquisition, device‑to‑cloud data handling, cloud batch inference, and integration testing when those are the priority. The eventual product is intended to operate as regulated medical‑device software. You will join and complement an existing ML engineer who owns important detector, data, and productization work. This role adds senior technical review, mentorship, and shared ownership. It is not a replacement role and does not include people management by default. Level may flex to Staff for a candidate who has demonstrated technical direction across multiple related systems.

What You Will Do

• Take technical responsibility for computer vision capabilities from problem definition and benchmark design through algorithm or model development, integration, and quantitative evaluation as a tested product component.
• Develop, train, and deploy models for sperm detection, multi‑object tracking, motility measurement, and image‑quality assessment using PyTorch and production‑grade frameworks.
• Design evaluation systems with subject‑disjoint dataset splits, versioned model artifacts, clinically relevant metrics, and error analysis that separates detector, tracker, reconstruction, and sample‑quality failures.
• Choose between classical image processing, deep learning, and combined approaches based on measured results rather than current trends.
• Define annotation requirements, labeling specifications, and ground‑truth standards; work with clinical collaborators and labeling partners to build and maintain training and test datasets.
• Carry models through GPU profiling, batch inference optimization, code review, testing, and deployment, not just to a working prototype.
• Contribute to device acquisition software, device‑to‑cloud data handling, cloud batch workers, APIs, and integration testing when those are the priority.
• Establish repeatable training workflows: versioned configurations, experiment tracking, model provenance, and traceable performance results.
• Use coding agents and engineering automation to accelerate implementation, testing, investigation, documentation, and traceability work, reviewing generated changes critically and verifying behavior before accepting them.
• Review experiments and code from the ML team; work directly with a developing ML engineer and provide a reliable technical sounding board.
• Investigate failures in dense clusters, cell crossings, debris, drift, blur, and reconstruction artifacts, and identify which part of the system is responsible.
• Work from incomplete specifications: define a practical first step, identify the decisions that require input, and produce a result that can be reviewed and built on.

Required Qualifications

• Demonstrated experience developing and shipping computer‑vision or image‑processing systems used in a production, clinical, industrial, scientific, or other real‑world operational environment, with responsibility beyond offline model development (integration, testing, deployment, monitoring, or sustained operation). Relevant areas include object detection, multi‑object tracking, segmentation, registration, motion estimation, video analysis, or image‑quality assessment.
• Strong Python skills and hands‑on production experience with PyTorch (or a comparable deep‑learning framework).
• Sound experimental practice: dataset splitting, metric selection, ablation studies, reproducibility, and error analysis, including how participant leakage, threshold reuse, and metric choices can misrepresent true performance.
• Software engineering fundamentals: typed interfaces, unit and integration testing, version control, code review, and technical documentation.
• Ability to work from an incompletely specified goal, determine what needs to be measured or built, scope a reasonable first step, and identify what decisions require broader input.
• A history of completing difficult technical work through integration, review, and delivery.
• Comfortable using coding agents as engineering tools while retaining technical accountability for the output.
• Clear written and verbal communication in English with technical and non‑technical collaborators.
• MS or PhD in Computer Science, Electrical Engineering, Biomedical Engineering, or a related field, or equivalent professional experience. Strong Senior candidates typically have five or more years of relevant experience; candidates with fewer years and directly applicable depth are also considered.

Preferred Qualifications

• Object detection for dense or small targets; oriented bounding boxes, rotated detection, Detectron2, MMDetection, or equivalent.
• Multi‑object tracking: data association, Kalman filtering, Hungarian assignment, optical flow, track stitching, identity continuity, or appearance‑based re‑identification.
• Medical imaging, digital pathology, microscopy, cytology, or laboratory automation, with familiarity with limited ground truth, inter‑annotator agreement, and measurement‑error analysis.
• ML infrastructure: experiment tracking (Weights & Biases, MLflow), model versioning, Dockerized training, cloud batch inference (GCP Batch, Vertex AI, or AWS), and GPU profiling.
• Regulated medical or safety‑critical software: IEC 62304, FDA SaMD, design controls, or analytical validation.
• Linux device software: camera or sensor acquisition, video handling, data upload under poor connectivity, or inference on constrained hardware.
• OpenCV, NumPy, SciPy, ONNX, TensorRT, or CUDA performance analysis.
• Robotics, autonomous systems, video analytics, or quantitative measurement where dense scenes, occlusion, and temporal consistency matter.
• Vision Transformers (ViT, DINO, DINOv2), self‑supervised pretraining, or foundation models applied to scientific or medical images.
• Statistical validation: repeatability and reproducibility studies, method comparison, Bland‑Altman analysis, or sensitivity and specificity estimation.
• Experience mentoring less‑experienced ML engineers.

What Success Looks Like

• Primary responsibility for one or more prioritized CV capabilities, such as detection, identity tracking, motility analysis, autofocus, image quality, or their evaluation systems.
• A core component integrated into the product architecture with a stable typed interface, unit and characterization tests, integration coverage, versioned models and configuration, and reproducible benchmark results.
• The component exercised across clean and difficult cases: dense fields, cell crossings, debris, poor focus, and low‑count samples.
• At least one substantive iteration that improves a meaningful product endpoint, not only an offline proxy metric.
• Major failure modes and operating limits identified and addressed.
• Improved review quality, test coverage, and reproducibility for the existing ML team.

Our Technical Environment

Our current detector is an oriented RetinaNet implemented in PyTorch with Detectron2 rotated‑box operations. Full stack: Python, PyTorch, Detectron2, OpenCV, GCP Batch, Vertex AI, GCS, Docker, GitLab CI, and a modular monorepo. We choose between classical and deep approaches based on measured evidence, document failure modes, and consider an experiment complete when it produces a reproducible result, a clear decision, or maintainable code.

Why This Role

Hard real‑world vision problems: dense moving‑cell video, holographic phase imagery, and a product spanning a physical scanner, cloud analysis, and a clinical workflow. A direct connection between your algorithms and measurements that influence clinical decisions. Technical responsibility from scientific question through product delivery. A small team where engineering practices are still being established and your contributions shape them. The opportunity to build the CV and ML foundation of a regulated medical product.

Location & Hours

Fully remote from Poland. The team works on US Eastern Time with flexible hours; we ask for reliable overlap with core Eastern Time hours (afternoons/early evenings in Warsaw) for reviews, pairing, and clinical‑collaborator meetings. B2B contract. Flexible paid time off.

illumicell AI is an equal opportunity employer. If you require a reasonable accommodation during the application or interview process, please let us know when you apply.