Making ML attack resilient

Machine Learning are not robust against adversarial attacks. We conducted approaches to improve that.

Over a three-year period, we worked closely with one of Russia's largest banks to successfully establish a robust AI monitoring system. This system operates with precision to swiftly address system failures by pinpointing their root causes and making well-informed decisions. The model expanded the scope of forecasted indicators to a total of 50,000, achieving an 85% reduction in false positive alerts. It also delivered a 20% improvement in precision compared to previous methodologies.

Led a team to develop an AI monitoring system that was capable of detecting anomalies in various business indicators. By closely working with one of Russia's largest banks over a span of 3 years, we designed and implemented an advanced backend system dedicated to the detection of anomalies across a wide spectrum of critical business metrics. This system significantly decreased the time and effort previously invested in manual incident detection and enhanced operational efficiency.

I coordinated a team in the development of two distinct automated incident detection models, which yielded substantial improvements in incident detection times, reducing them by an impressive 5 to 15 minutes. These models played a pivotal role in streamlining processes and optimizing the bank's operations.

• Developed AI assistance leveraging user data for diverse scenarios including question answering and business intelligence.
• Managed stakeholder relationships with service owners and external clients to align needs and expectations.
• Collected and documented high-precision business requirements focusing on API and AI model parameters.
• Conducted data analysis to identify trends and insights for AI-driven business decisions.
• Aligned design solutions with project goals through collaboration with UX designers.