Malik Gazizullin

Fine-tuned a state-of-the-art Vision-Language Model (VLM) for use in a multimodal virtual assistant. This enhanced performance on real-world image-query tasks and improved user interaction quality.

Led development of an online service processing over 200 news sources to increase maps quality, reducing cartographers’ manual work time by 60%. Spearheaded a company recognition project from vehicle images, achieving a recall rate of 15%, outperforming internal benchmarks of 8-12%.

Led a team to optimize service efficiency with concurrent data processing and SOTA ML models, achieving a 100x increase in daily throughput and a 24x reduction in latency. Directed the expansion of the computer vision pipeline, boosting recall from 15% to 49% through new data sources, modalities, and advanced algorithms, driving significant cost reductions.

Led development of deep neural network models, deploying segmentation algorithms over 13,000 km of road surfaces in aerial imagery. Pioneered advanced neural networks for fast detection of road markings in satellite/aerial data, increasing marking efficiency by 20%.

Developed and deployed deep learning models that significantly enhanced unwanted content recognition in real time, improving accuracy and recall by over 20% and tripling precision in weapon classification. Spearheaded data collection and model training for image classification, pioneering NLP and classical ML models to enhance gif and video classification, and led an OCR project achieving notable

Directed the management and development of crucial linear and logistic regression and gradient boosting models, resulting in an 8% accuracy boost in the primary conversion prediction model and significant revenue increase. Spearheaded the development and integration of training, deployment, and inference modules, enhancing team workflow efficiency and scalability.

Spearheaded barcode recognition advancements, optimizing latency by an impressive 13%. Conducted pioneering research into Neural Architecture Search (NAS) on Pytorch, boosting SOTA model performance by 2% in precision and 1% in F1 score in object detection and segmentation tasks.

Completed a Master's degree in Computer Science. Focused on advanced topics in the field, with a thesis titled 'Cascade Neural Network for Barcode Detection'.

Participated in the Machine Learning Developer program, specializing in Computer Vision (Classical and Deep learning, 3D), Reinforcement Learning, and Natural Language Processing. Acquired practical skills in developing and deploying advanced machine learning solutions.

Earned a Bachelor's degree in Computer Science. Conducted research for the thesis on 'Neural Architecture Search for barcode detection task'.