NanoFIDSign in

About NanoFID

Oleh Lahoda0009-0000-4550-3603

The researcher

My name is Oleh Lahoda. I am a 17-year-old student at the Dublin School, where I attend on a full-tuition scholarship. In the summer of 2026, I will study machine learning at the Stanford Pre-College Summer Program, where I have also received a full-tuition scholarship to deepen my background in modern ML systems.

Beyond my school curriculum, I have completed independent coursework in computer science and machine learning, including:

  • Harvard CS50: Introduction to Computer Science
  • MIT 6.0001: Introduction to Computer Science and Programming in Python
  • MIT 6.036: Introduction to Machine Learning

NanoFID is an independent research platform I designed and built to study the inner mechanics of facial-recognition systems and the human factors that shape their behavior.

Why this project exists

Face identification is one of the most widely deployed machine-learning technologies in everyday life. Yet for most people who interact with it daily, FaceID remains a black box. Few users can articulate how a model is trained, why it sometimes fails on faces it has already seen, or why two visually different people can register similar confidence scores to the same model.

I built NanoFID to open that black box. The platform allows any user to train a small face-recognition model on their own photographs, observe how it behaves under varying conditions, and review the same diagnostic signals that ML researchers use to evaluate production systems. The goal is not only to collect data but to make the training process itself legible.

What I hope users take away

Beyond the research outputs, I want every person who uses NanoFID to leave with a clearer understanding of how face recognition, and machine learning more broadly, actually works. Mathematics is not magic. Embedding spaces are not mystical. Confidence scores are not certainty. In an era where AI systems make consequential decisions about ordinary people, the public deserves the vocabulary and intuitions to understand those systems on their own terms.

If NanoFID succeeds, it will not only contribute new datasets to the bias-research literature. It will leave its users with a more accurate, more confident, and more critical understanding of the algorithms that increasingly shape their lives.

What NanoFID measures

The platform investigates three intertwined questions.

Algorithmic bias and fairness. When a face-recognition model is exposed to faces across different demographic groups, where does its accuracy degrade, and by how much? NanoFID logs anonymized confidence and identification metrics across user-contributed datasets to surface bias patterns at scale.

Generalization gaps in machine learning. A model that performs well on its training data does not necessarily generalize. NanoFID explicitly measures the gap between training-set accuracy and held-out test accuracy on every model trained, allowing users to see, often in real time, how brittle or robust their personal model is.

Privacy and consent in biometric systems. All photographs uploaded to NanoFID are stored privately, processed with the user's explicit consent, and deleted after a finite retention window. The platform itself is a study in how biometric research can be conducted with transparent, user-aligned data practices.

Further reading

If you wish to read the full research paper accompanying this project, including the engineering decisions made during its development, please refer to my ORCID record: 0009-0000-4550-3603.

Get in touch