• Implemented an end-to-end heavily parallelised Auto-ML pipeline to automate data analysis and feature predictions of customer data with Neural Networks (e.g. LightGBM, XGBoost and Ray).• Collaborated with clients to develop Proof of Concepts (POCs) and demonstrations powered by AI.• Developed project proposals.• Bridged technical details to non-technical stake holders.• Managed the product integration process on the customer’s side.Stack: Kubernetes, Docker, Ray… Show more • Implemented an end-to-end heavily parallelised Auto-ML pipeline to automate data analysis and feature predictions of customer data with Neural Networks (e.g. LightGBM, XGBoost and Ray).• Collaborated with clients to develop Proof of Concepts (POCs) and demonstrations powered by AI.• Developed project proposals.• Bridged technical details to non-technical stake holders.• Managed the product integration process on the customer’s side.Stack: Kubernetes, Docker, Ray, Prefect, MongoDB, Pandas, Numpy, Scikit-Learn, Python, Swift Show less

- Developed and implemented an AutoML pipeline using AWS SageMaker Pipeline, S3 and Lambda frameworks, resulting in a 16% improvement in time-series data forecasting accuracy.- Trained, tuned, and deployed XGBoost models for time-series forecasting, achieving 96% average accuracy.

• Implemented an end-to-end heavily parallelised Auto-ML pipeline to automate data analysis and feature predictions of customer data with Neural Networks (e.g. LightGBM, XGBoost and Ray).• Collaborated with clients to develop Proof of Concepts (POCs) and demonstrations powered by AI.• Developed project proposals.• Bridged technical details to non-technical stake holders.• Managed the product integration process on the customer’s side.Stack: Kubernetes, Docker, Ray… Show more • Implemented an end-to-end heavily parallelised Auto-ML pipeline to automate data analysis and feature predictions of customer data with Neural Networks (e.g. LightGBM, XGBoost and Ray).• Collaborated with clients to develop Proof of Concepts (POCs) and demonstrations powered by AI.• Developed project proposals.• Bridged technical details to non-technical stake holders.• Managed the product integration process on the customer’s side.Stack: Kubernetes, Docker, Ray, Prefect, MongoDB, Pandas, Numpy, Scikit-Learn, Python, Swift Show less

Utilized a SOTA Transformer-type Language Model (BERT) to successfully replicate the original data distribution, resulting in a convenient and accurate snapshot of an anonymized user data. That improved trained Language Model Perplexity by 20% in experience replay for incremental learning, and eliminated the risk of customer data loss.Stack: Python, Tensorflow, Pytorch, Cuda, Pandas, Numpy, Scikit-Learn

- Developed an innovative algorithm to enhance automatic speech recognition for Attention-based Encoder-Decoder models, resulting in a 6% reduction in Word Error Rate, surpassing the current SOTA Language Model interpolation method. - Trained, evaluated and fine-tuned deep learning models (e. g. AED, Transformer) achieving 5.7 WER score on Librispeech test set with AED model.

iOS software developer

Feature data processing. Implemented a mathematical algorithm of an invariant rotation of points clouds for restoring objects from a 2D picture as a data augmentation method for Computer Vision purposes.

Research assistant in the laboratory of nano-lithography. Implemented a computer simulation model of a neutral filter for EUV range in Python for spectroscopy and nano-litography experiments.