Denis Surnin Email & Phone Number

Building AI solutions for a long-lasting change

Multi-Task Bert Model• Successfully developed a Multi-Task Learning (MTL) BERT model, achieving up to 6% improvement in accuracy for NER and Intent labeling tasks through joint training optimization.• Extended the BERT architecture with custom output layers for enhanced NER and Intent labeling.• Leveraged Weights & Biases for advanced monitoring of training metrics, optimizing model parameters in real-time to reduce overfitting by 9%.• Ensured the codebase was production-ready, adhering to industry-standard coding practices and implementing comprehensive tests, reducing potential deployment errors.• Streamlined model training and deployment using PyTorch Lightning, successfully deploying the demo model to HuggingFace Spaces for evaluation and interaction.Project on Mitigating Catastrophic Forgetting• EWC (Elastic Weight Consolidation): Implemented EWC to maintain 30% accuracy on initial tasks after subsequent training, identifying critical insights into Fisher's matrix impact on performance. Experimented extensively with the ewc_lambda parameter, adapting to dataset specifics and contributing to understanding parameter tuning's effect on model robustness.• L2F (Learning without Forgetting): Advanced dynamic architecture manipulation skills by adapting a model for incremental task learning, achieving 32% accuracy on initial tasks. Pioneered weight transfer techniques and selective parameter freezing to enhance model adaptability and task memory retention.• MCL (Meta Continual Learning): Applied MCL across divided tasks within the FashionMNIST dataset, securing accuracies of 32% and 30% for the first two tasks, respectively.• SI (Synaptic Intelligence): Engineered a surrogate loss function based on parameter importance, achieving 15% accuracy on foundational tasks. This process highlighted the nuanced balance between parameter significance and model adaptability.

Izhevsk, Udmurtia, Russia

• Implemented advanced NDT inspections, leveraging ultrasonic, magnetic particle, and dye penetrant testing techniques, enhancing material integrity assessment accuracy by 10%.• Expertly analyzed NDT results, identifying critical defects and structural weaknesses, which informed the development of targeted corrective actions, reducing potential safety risks by 15%.• Crafted and delivered comprehensive technical reports, ensuring clear communication of inspection findings to stakeholders.

Perm, Russia

• Engineered and drafted innovative bridge slipway designs, supporting the construction of 150-meter slipways and optimizing metal use, leading to a 10% cost reduction and 15% increase in construction efficiency.• Oversaw critical construction processes, including blast cleaning and structural installation, ensuring adherence to best practices and enhancing project quality compliance by 20%.• Prepared and submitted technical reports like accomplishments, progress, and other matters concerning the ongoing project• Performed rigorous daily onsite inspections, enhancing technical and safety standards compliance, directly contributing to a 10% reduction in safety incidents.

Izhevsk, Udmurtia, Russia

• Pioneered the design and engineering calculations for a rain-screen reconstruction project, delivering ahead of schedule and ensuring enhanced building protection and energy efficiency.• Led the collaborative design effort for building reconstruction and aesthetic enhancements, resulting in a 10% improvement in project tender competitiveness and stakeholder satisfaction.• Delivered crucial design and drafting support, streamlining project phases and contributing to a 5% reduction in design turnaround times.• Formulated detailed engineering calculations and reports, establishing a robust foundation for structural integrity and design optimization in rain-screen projects.

Master Of Engineering - Meng, Engineering/Industrial Management

Bachelor Of Engineering - Be, Civil Engineering