Artsiom worked on quantization and model optimization features for edge AI workflows, contributing to the google-ai-edge/LiteRT and google-ai-edge/ai-edge-quantizer repositories. He developed a configurable option in LiteRT’s TensorFlow Lite calibration pipeline to selectively disable per-channel quantization for dense layers, giving users finer control over the trade-off between model accuracy and inference performance. In ai-edge-quantizer, he implemented end-to-end Mean Squared Error–based quantization materialization for convolutional layers and integrated these functions into the algorithm manager, streamlining quantization workflows. His work leveraged C++, Python, and TensorFlow Lite, demonstrating depth in quantization techniques and practical integration into production pipelines.