Nikita Popkov

M.Sc. Computer Science — Applied ML & Physical Systems

Kassel-based researcher focused on machine learning for physical systems, time series, and real-time experimentation.

Experiences

research stay

European XFEL Research Stay in Hamburg

European XFEL, Hamburg

Implemented and tested AI models directly on the linear accelerator, built a live experimental data pipeline, and delivered real-time predictions in an international and -disciplinary team. Stayed for three weeks and participated in two beamtimes.

pythonpytorchkarabonumpy
research collaboration

TRANSALP — Laser Pulse Prediction

BMBF Forschungsprojekt

Developed machine learning models to predict laser pulse characteristics and supported quality assurance across an international research team.

pythonpytorchpandasnumpy
research

Particle Detection Framework

University of Kassel

Co-developed a physics experiment control framework and trained AI models to detect magnetic particles and track their trajectories under a microscope.

pythonpytorchdeeptrackdeeplaytrackpytango
research project

SylasKI — Synthetic Load Time Series Generation

BMBF Forschungsprojekt

Built a custom diffusion-based model for synthetic household load time series generation and evaluated it on German and UK energy datasets.

pythonpytorchpandassql

Education

thesis

M.Sc. Thesis: Closed-Loop Experimentation for Physical Applications

Nikita Popkov

University of Kassel

Research on iterative, experiment-driven model improvement for physical systems, with emphasis on feedback stability and real-data constraints. Developed a TANGO based device network structure that allows real-time feature extraction and experimental analysis.

machine learningclosed-loopautomated experimentation
thesis

B.Sc. Thesis: Time Series Denoising with Diffusion Probabilistic Models

Nikita Popkov

University of Kassel

Investigated denoising diffusion probabilistic models for time series, with application to energy load data cleaning and synthetic generation. Building own DDPM model architecture, training the model on a set of german household data and evaluating it against a set timeframe.

machine learningdiffusion modelstime series

Personal Projects

Double Ratchet X3DH Encrypted CLI Chat

A terminal-to-terminal encrypted chat protocol implementing X3DH key agreement and the Double Ratchet algorithm for state-of-the-art message encryption.

PythonCryptographyX3DH

Finance-Bot

Automated data retrieval and analysis for financial products. Processes daily price movements for information extraction, generates forecasts from historical data, and autonomously executes buy/sell decisions based on extracted signals.

PythonWebSocketsPandas

Photos to 3D-Mesh

Photogrammetry pipeline that converts photos of objects into 3D models using Meshroom, post-processed in Blender, and printed as miniatures on a 3D printer.

MeshroomBlender3D Printing

Smarthome

A collection of projects to make the home smarter — an e-ink display for calendar, weather, and shopping lists; controllable lighting; and a mini ESP display for videos and images; as well as setting up a home assistent via hermes agents.

ArduinoRaspberry PiESP32Blender3D PrintingHermes Agent