Mikhail Filippov

Architected an end-to-end LLM-powered matching pipeline combining deterministic logic, vector search, and conditional multi-model inference. Reduced LLM usage by approximately 60% through severity-based decision logic, semantic caching, and early-exit strategies. Improved end-to-end latency by approximately 67% using fully async batch processing and optimized concurrency. Built a backend supporting deal discovery, price alerts, smart shopping lists, analytics, and AI-powered explanations.

Led migration from monolith to microservices, boosting deployment speed by 40% and system resilience. Designed an event-driven architecture with RabbitMQ for fault-tolerant async communication. Built an internal chatbot, cutting support requests by 35%. Developed an AI assistant for financial analysts using LLMs and databases, enabling natural language queries and reducing analysis time by 40%.

Developed automated tools to identify company tickers, retrieve live market data, and collect financial reports, providing analysts with fast, accurate insights. Streamlined data extraction using Selenium WebDriver, improving efficiency and freeing up team resources. Optimized PostgreSQL queries, reducing execution time by 50%.

Led the development and upkeep of an intuitive website builder tailored for non-profits, delivering smooth user experience and high system performance. Introduced multi-tier caching with Redis, CDN, and browser storage, cutting average page load time from 3.2 to 0.8 seconds.

Directed the design and implementation of a corporate EDMS handling over 15,000 daily transactions. Integrated OAuth 2.0 and JWT for secure, unified access across multiple portals. Built core components using DDD and microservices architecture, increasing document processing speed by 35%.