Engineering · Paid internship · 1 opening
Computer Vision Engineering Intern
Model Training & Pipeline
Help turn soccer video into useful training insight. Work across the practical side of computer vision: the data, the training runs and the detection pipeline.
What we’re building
KickMap uses phone video of soccer kicks to help players understand the strike: foot-to-ball contact and the resulting ball speed, launch angle, spin, curve and trajectory. The goal is objective feedback that helps individual players—and players across a roster—train with greater purpose.
The engineering challenge is to turn ordinary footage into trustworthy insight, not just produce a convincing-looking detection. Better ball contact changes everything.
Why this is technically interesting
Foot-to-ball contact lasts milliseconds. Motion blur and occlusion can hide the moment that matters most. Our footage comes from backyards and sidelines, with changing light, weather and camera angles—not a controlled lab.
Small detection errors can compound into misleading downstream physics. That makes high-quality, targeted labeling enormously important: the right examples and consistent annotations can matter as much as another training run.
Three connected workstreams
You’ll work across the data, the model and the production pipeline, with Adam helping you connect decisions in one stage to results in the next.
1. Video labeling & dataset curation
- Tag ball and player detections, foot-to-ball contact frames and kick outcomes.
- Refine labeling guidelines so annotations are consistent and repeatable.
- Audit labeling errors and curate edge cases, including difficult footage the model gets wrong.
- Keep datasets versioned so training and evaluation results can be traced back to their source data.
2. Model training & evaluation
- Run training scripts and configurations, and watch loss, validation results and GPU usage.
- Compare runs using held-out evaluations rather than judging improvement from training loss alone.
- Record experiments: what changed, how the model performed and what you learned.
3. Production detection pipeline
- Monitor detection confidence, failures and latency in the production pipeline.
- Triage problems: is the issue in the source footage, the model or the code?
- Build dashboards and alerts that make problems easier to spot and investigate.
- Close the feedback loop with Adam and the team, turning production failures into better data, experiments and pipeline changes.
An honest note about the day-to-day
Expect significant repetitive labeling at first. Careful, consistent work is central to this role—not something to rush through before the “real” engineering begins.
Over time, you’ll take on increasing autonomy in training and pipeline work as you build context and confidence.
How we use AI
AI-assisted coding, labeling and reviews are encouraged. But every output needs critical verification. You should be able to explain your decisions, check the results and recognize when an AI suggestion is wrong or incomplete.
What you bring
Minimum qualifications
- Currently enrolled in computer science, data science, engineering or a related field.
- Python experience.
- Some familiarity with machine learning or computer vision, including training, validation and loss.
- Comfort with Git and the terminal.
- The ability to do careful, repeatable, detail-oriented work.
- Clear communication, curiosity and a willingness to ask questions.
Nice to have—not required
- PyTorch and object detection or tracking experience.
- OpenCV or ffmpeg for working with video.
- Labeling tools or dataset curation experience.
- Cloud GPU workflows.
- SQL or dashboards for exploring and monitoring data.
Soccer knowledge is welcome, but optional. You don’t need to be a player to contribute.
What you’ll gain
Real end-to-end ML engineering experience: understanding footage, building datasets, evaluating models and monitoring what happens in production.
- A weekly 1:1 with Adam Spaulding and feedback on your work.
- Structured goals that connect your day-to-day work to model and pipeline improvements.
- The opportunity to demonstrate model improvement through recorded experiments and evaluations.
- A possible reference based on your work.
This internship does not promise subsequent full-time employment.
Mentorship & working arrangements
Your mentor is Adam Spaulding.
There is one opening, paid $17–$21 per hour. This is a remote internship with occasional travel to Dover, New Hampshire. Start and end dates are flexible around school, and hours are arranged around classes. You’ll need a few hours of overlap with US Eastern business hours.
Applications
Apply by email
Send your application to adam.spaulding@gmail.com with the subject “CV Intern — [Your Name]”. Include:
- Your resume as a PDF. No cover letter needed.
- A link to a project you built or contributed to, with a note explaining your personal contribution.
- 2–3 sentences on what you hope to learn.
- 1–2 sentences with evidence of being careful and consistent with detail-oriented work.
Applications reviewed on a rolling basis.
Equal opportunity & accommodations
We consider candidates without regard to race, color, religion, sex, sexual orientation, gender identity, national origin, disability, veteran status, or other protected characteristic.
If you need an accommodation during the application process, please let Adam know by email.
Better ball contact changes everything.