KM

Open to Data Science roles

Kseniia Melnikova

Network Traffic Analyst Data Scientist

I came to data science from bank cybersecurity, where a model runs in production and its mistakes become real incidents. That taught me to value clean data, honest validation and results that hold up outside the notebook.

01whoami

Signal in the noise

02experience

Three years in bank cybersecurity

    03education

    University and further training

    04skills

    Tools I can put to work

    No self-assigned percentages. Each item says where I used it and what the work involved.

    05projects

    Selected project

    06interactive

    The Anomaly Lab

    This is a place to play with anomaly detection and see what a single score usually hides. The detector first learns what normal traffic looks like, then decides which new events seem suspicious. Move the threshold down and it catches more anomalies but raises more false alarms. Move it up and the screen gets quieter, while more attacks slip through. The simulation knows the correct answer for every event, so precision, recall and false-positive rate are real. Try choosing a threshold or fooling the detector.

    Interactive anomaly-detection simulation. Normal and anomalous events are scored against a reference baseline. Use the controls below to inject events and inspect detector metrics.

      Click the field: add a normal event · Shift+click: add an anomaly

      Lower threshold: more alerts and higher sensitivity. Higher threshold: fewer alerts and lower sensitivity.

      Baseline

      Teaching experiment: every evaluated event, including anomalies, joins the baseline. A repeated attack gradually stops being detected.

      Detector metrics

        07off the clock

        Life outside the notebook

        Away from data, I travel to get under the skin of a place — its language, its habits, its food. I love river rafting, camping and hiking, and I photograph it all on my own camera. Before I moved, I was into postcrossing: trading handwritten postcards with strangers around the world.

        08contact

        Let’s talk

        I’m interested in problems where training a model is only half the job. The other half is understanding the data, testing the result honestly and explaining what it means. If that sounds relevant to your team, or you just want to say hi, drop me a line.

        Email or Telegram, whichever is easier.

        ↑↓ navigate · run · esc close