Project: Website and environment for a neural network distillation competition (Transformers,RNN) for NLP tasks
Tasks: Optimization and distillation of neural networks (LSTM, Transformers )
Developed pipelines for recurrent neural network training and data processing for NLP tasks
Deploy and evaluate model quality using (MLFlow)
Developed API (FastAPI) for user interaction with models and data (MySQL)
Technologies: FastAPI, MLFlow, PyTorch, MySQL
Team: 1 full stack, 2 ML, 1 project manager, 2 academic researchers
Methodology: agile (scrum)
Result: The project was completed in accordance with the TOR and put into production. According to the results of the competition a scientific article was published
Duration: about 1 year


Project: A ‘’Smart Glove' device” capable of detecting gestures and recognizing letters
written in the air for VR and AR
Objectives: Selection of sensors and design of the “”Smart Glove“” device
✔️ Collect and then analyze multi-user datasets
✔️ Developing a neural network (CNN and LSTM) to handle time series
✔️ Utilizing compression and distillation of neural network models
✔️ Working with ChatGPT and LangChain to create a chatbot on their database
Technologies: MLFlow, Tensorflow Lite, Tensorflow, OpenVino, ONNX, LangChain
Team: 1 engineer, 1 ML , 1 project manager, 1 research scientist
Methodology: agile (scrum, kanban)
Result: The first stage of the project was completed according to the ToR and is fully functional
Duration: 7 months

Project - by transcribing audio into text using third-party APIs
After translating them into vector using embedding and using RAG technology to find the necessary vector and text Tasks -Based on this information generated a prompt for LLM and communicated with the user
Moreover I connected xls tables to LLM via google sheet API
Duration - about 2 months and now still in work