Middle fullstack ML developer

О себе

---

На данный момент Middle fullstack ML developer.

Мои компетенции и опыт

Achievements:

 - developed and deployed a real-time object detection system using YOLO and OpenCV with custom bounding box processor, achieving 99% accuracy;

 - implemented a custom conditioned U-Net architecture for synthetic image generation for defects segmentation tasks, improving precision up to 99.5%;

 - built and deployed RESTful APIs using FastAPI to serve ML models in production environments;

 - conducted research on diffusion models for generative tasks, contributing to cutting-edge AI advancements.

TECHNOLOGIES

Languages:Python, C, C++, Bash, Matlab, Mathematica, asmTechnologies and tools:PyTorch, OpenCV, sklearn, onnx, tensorflow, keras, PIL, Albumentations, Dlib, numpy, pandas, matplotlib, seaborn, Flask, FastAPI.Neural Networks:CNN, RNN, ResNet, U-net, transformers, ViT, ResNet, VGG, Yolo, SSD, Faster R-CNN, Mask R-CNN, diffusion models.DBMS:MongoDB, SQLite, chromadb, pinecone, Postrgress, supabaseDevOps and Clouds:ClearML, DockerTesting:Unittest (python)Version Control:Git, svm, mercurial

Huawei SPBRC Russia Key projects engineer March 2022 – June 2025

3D GenAI for Cloud Gaming – creating new and adopting sota technologies for Cloud Gaming department

Personal results:

New technology was developed for controlled high-quality diffusion-based 3D generation;

Technologies overview for 3D generation;

Gathering requirements and presenting results to the stakeholders.

RAG for internal purposes – indexing and advanced search over a pile of internal documents

Personal results:

Optimizing embedder, chunker, LLM finetuning & quantization for thousands documents scale RAG system.

Intellectstack Ltd. CTO, founder April 2020 – December 2023

Various projects for commercial customers

Personal results:

Leading a group of engineers;

Gathering requirements and presenting results to the stakeholders;

Full cycle development.

 

Activetest Group Ltd. Deputy Technical Director June 2020 – March 2022

Equipment and machinery for the mining, energy, oil and gas industries (manufacturing). 

Personal results:

Management of software development processes for machine control, data acquisition and processing, visualization, and report generation;

ML & CV Technologies overview;

Extensive use of synthetic data as a primary source of training data.

 

Protei STC Ltd. Tech lead June 2014 – September 2019

Computer vision for Safe City Project.

Personal results:

The developed system is characterized by low requirements for existing video equipment and high ease of configuration, while maintaining high recognition quality. Easy integration into the existing infrastructure allowed the customer to achieve their goals, significantly saving on the cost of system implementation.

Video analysis of road conditions, detection of traffic violations.

Personal results:

The use of modern methods made it possible to achieve the required quality of detection and tracking on server systems optimized with OpenVINO without the use of GPUs, which reduced the cost of the solutions provided to the customer.

Call center usage forecast.

Personal results:

The solution enabled the customer to reduce the number of simultaneously working operators while maintaining the required level of service, thereby lowering costs.

Planning, development, testing and maintenance of a fully automated multi-camera system for industrial video analytics.

Personal results:

High-load system of ~5000 cameras (~300 per server);

Sparse video stream: The analytics do not process a continuous video stream but instead work with individual frames requested at intervals, significantly reducing network bandwidth and server requirements.
 

Texas A&M University. USA: PostDoc August 2012 – December 2013

Computer simulation of physical processes, programming of Varian spectrometers, overexpression, isolation, purification, and crystallization of proteins.

 

Miami University. USA: PostDoc November 2009 – August 2012

Development of a new method for measuring the depth of insertion and position of proteins in a lipid membrane using solid-state NMR (paramagnetic relaxation effect) and controlled incorporation of a spin label into the protein

 

University of Southampton, UK: PostDoc October 2007 – October 2009

Programming of Bruker and Varian spectrometers, working with experimental samples using cryogenic MAS technology, NMR of proteins, computer simulation of spin dynamics in SIMPSON.

 

Bundesanstalt für Materialforschung und -prüfung (BAM), Berlin, Germany: PhD September 2004 – September 2007

Study of biomineralization processes in bones and nacre. The key role of chondroitin sulfate in the biomineralization of calcium phosphates in bones has been discovered.

EDUCATION, DIPLOMA AND ACHIEVEMENTS

2008 Leipzig Universität, Leipzig, Deutschland: PhD: Physik und Geowissenschaften, Physics

2004 Saint-Petersburg State Polytechnical University: Master: Physics&Mechanics dep. Experimental nuclear physics.

2021 City Business School, MBA General Intensive

2015 Nvidia CUDA programming

2013 Learning from Data, Caltech, ML

2013 Web Development, Udacity, Web programming

2013 Machine Learning (Andrew Ng) Coursera

2013 Hadoop and MapReduce, Udacity, BigData

2013 Databases, Coursera / Stanford

2011 AI course (Sebastian Thrun) Udacity

 

PUBLICATIONS list sent per request