Computer Vision Engineer

About this role

SumerSports is a leading football intelligence technology company that specializes in providing an innovative suite of products for football fans and NFL clubs. We are a collection of executives, engineers, data scientists, and visionaries from NFL clubs, technology startups, finance, and academia. Our data-driven platform empowers teams with insights and tools to make informed decisions within salary cap constraints. The platform also serves the NCAA, offering insights around the transfer portal and more. What sets us apart is our unique blend of big tech talent, data scientists, and former NFL personnel, who have a combined 600+ years of NFL experience. Our domain knowledge is augmented by AI and machine learning technologies to create a unique view into many aspects of Football. We’re hiring a hands-on Computer Vision Engineer to build and improve sports video intelligence models—detection, tracking, pose, event understanding, and multi-view reasoning. You’ll spend most of your time on CV research + applied modeling (experiments, architectures, training, evaluation), and partner with data/platform teammates to ensure your work can ship reliably. This role is CV-first. A bend toward scalable pipelines / MLOps is a plus, not a requirement. Level (mid vs senior) depends on scope ownership and how independently you can drive results. Responsibilities CV Modeling & Experimentation

Build and train CV models for sports video: player/ball detection, multi-object tracking, pose/keypoints, event/action recognition, identity association (re-ID).

Own the experimentation loop: hypotheses → ablations → error analysis → measurable improvements.

Design and maintain evaluation: task-appropriate metrics (e.g., MOT metrics, keypoint accuracy, event precision/recall), dataset slices, and failure taxonomy.

Improve data efficiency: augmentations, sampling strategies, handling label noise, weak/self-supervision where helpful.

Prototype and iterate on modern architectures (e.g., transformer-based detection/tracking, temporal models, multi-task setups).

Research that Ships

Collaborate on dataset + labeling design: formats, schemas, tooling, versioning.

Help productionize models: packaging, batch/stream inference patterns, throughput/latency tradeoffs, robustness checks.

Add lightweight quality gates: reproducibility, automated eval, regression detection

Qualifications Must-have:

Strong applied CV experience with hands-on model development (not just running existing repos).

Solid PyTorch skills: training loops, debugging, data pipelines for vision workloads, DDP basics.

Comfort with video CV fundamentals: occlusion, identity switches, temporal consistency, calibration, domain shift.

Strong Python engineering and a bias toward measurable outcomes.

Nice-to-have (Bonus):

Sports video CV or adjacent domains (multi-agent tracking, pose, crowded scenes).

Experience with video tooling (FFmpeg), efficient dataset formats (WebDataset/shards), or streaming/batching to GPUs.

MLOps/production experience: model packaging, CI for training/eval, serving (Triton/TorchServe), monitoring.

Benefits

MLOps/production experience: model packaging, CI for training/eval, serving (Triton/TorchServe), monitoring.

Competitive Salary and Bonus Plan

Comprehensive health insurance plan

Retirement savings plan (401k) with company match

Generous paid holiday schedule - 13 in total including Monday after the Super Bowl

Remote working environment

Generous paid holiday schedule - 13 in total including Monday after the Super Bowl

Originally posted on Himalayas