Maksim Kobelev Email & Phone Number

Budapest, HU

Budapest, Hungary

• Leading team of 8 developers at Semi-structured Documents Processing group
• Created solution design and led experiments of the transition from outdated light-weight NeuralNetwork architecture to new-fashion BERT-based NN that allows to use few-shot learning techniqueson-premise, requires less.
• Implemented passing prior hypotheses from general pre-trained Neural Networks to userdomain-specific trained “Fast-Learning” technology that increases extraction quality both on sharedfields subset and on user-specific.
• Led experiments of increasing documents clustering quality with a new clustering engine based onmulti-type documents fine-tuned Visual-Transformer as an advanced image feature extractor.Participated in design of.
• Increased quality of existing “Fixed-Forms” fields extraction technology by transition from LeastSquares Method to RANSAC algorithm to find the most suitable affine transformation of most reliablereference points from.

Budapest, Hungary

• Leading team of 3 developers at User-side ML team.
• Implemented automatic extraction mode for side-by-side tables recognition in “Fast-Learning”technology, based on classic ML techniques.
• Solution designing the transition from rule-based to multimodal document classifiers for on-premiseclients. Managing all works from idea to production ready solution.

Moscow, Russia

• Researched on combining pixel-wise segmentation models (trained on different tables) with widelyused NN and Classic ML baseline architectures for Table extraction from invoices. As a result,provided a methodology of.
• Researched on classifying words on receipt documents with Graph Convolutional Networks.Developed a unique data preparation procedure and implemented modified GCN-layers for modeltraining. This technique increased micro.

Moscow, Russia

Developing Ansible playbooks for automatic deployment of secure images on virtual machines.Automatic installing and configuring software e.g. NGINX, NextCloud etc, for secured bank datastorages.

Master'S Degree, Financial Technologies And Data Analysis, 7.9

Bachelor'S Degree, Applied Mathematics And Informatics, 7.8