About Aleksei Zhukov · Larch Tools

Who makes Larch Tools

I’m Aleksei Zhukov, a data scientist with a PhD and more than nine years of applied machine learning, mostly on raw and messy data: geospatial, time series, images and 3D point clouds. Larch Tools is where I build small tools for data that’s hard to open, tested on real files before they claim anything.

What I can help with

Area	What I do
Machine learning and deep learning	Models for images and 3D point clouds, time series and control problems, taken from raw data to validated results.Papers:Ensemble classifiers for power system security, 2017 Deep reinforcement learning for microgrids, 2019
LLM pipelines	Pipelines built on large language models that analyse data and automate routine work, with checks on what the model returns.
Geospatial data	Deep learning and data pipelines for geospatial and environmental data, maps and location history.Tools:Map Image Export Google Timeline Converter
Data and models that are hard to use	Models that won’t load after an upgrade, moved to your current stack with a check that they predict the same; old databases and exports turned into clean data, checked value by value.Tools:Joblib Inspector MDB Viewer Parquet Viewer

Area

What I do

Machine learning and deep learning

Models for images and 3D point clouds, time series and control problems, taken from raw data to validated results.Papers:Ensemble classifiers for power system security, 2017 Deep reinforcement learning for microgrids, 2019

LLM pipelines

Pipelines built on large language models that analyse data and automate routine work, with checks on what the model returns.

Geospatial data

Deep learning and data pipelines for geospatial and environmental data, maps and location history.Tools:Map Image Export Google Timeline Converter

Data and models that are hard to use

Models that won’t load after an upgrade, moved to your current stack with a check that they predict the same; old databases and exports turned into clean data, checked value by value.Tools:Joblib Inspector MDB Viewer Parquet Viewer

Contact

Email is best: say what you’re working with and what you need. For a model or file that won’t open, the versions and the exact error are enough; please don’t attach the file itself. If I need to see real data, we agree on an NDA first and I work inside your environment, so the data stays with you.

My PhD is on machine learning methods for power system security assessment, on a grounding of mathematics, statistics, optimization and algorithm design. I work end to end: data exploration and feature engineering, modelling, evaluation and deployment.

Selected work

Deep learning models (U-Net) for super-resolution and downscaling of geospatial data.
Automated detection of road defects from 3D point clouds and 2D imagery.
Reinforcement learning for controlling renewable energy systems under uncertainty.
Forecasting models for the planning and stability of power systems with renewables.
NLP text matching in the medical and geospatial domains.
LLM pipelines for data analysis and for automating routine work.

Selected papers

Ensemble methods of classification for power systems security assessment. Applied Computing and Informatics, 2017.
On-line power systems security assessment using data stream random forest algorithm modification. Studies in Computational Intelligence, Springer, 2017.
Machine learning techniques for power system security assessment. IFAC-PapersOnLine, 2016.
Deep reinforcement learning for energy microgrids management considering flexible energy sources. EPJ Web of Conferences, 2019.

More than 20 peer-reviewed papers in applied machine learning, all on Google Scholar.

Tools I use

Python with NumPy, pandas, SciPy and scikit-learn; large language models in data pipelines; working knowledge of PyTorch and TensorFlow; geospatial data tools; MATLAB and R.