Illia Stebelski

Built batch-inference engine on OpenAI Batch API, halving LLM token cost for document classification and embeddings. Made LangGraph chat agent reliable with tool-call budgets and caching, and kept AI document-review grid fast at 30M cells.

Led a team of 5 backend engineers in splitting a multi-tenant monolith into independently deployable services. Designed Kafka event streaming and CQRS communication, and ran services on Kubernetes with observability.

Cut a core processing job from 5 hours to 15 seconds using set-based SQL with CTEs. Built and operated 5 microservices for pharmacy stock, receipts, and financial reporting, and rebuilt the API testing framework to raise coverage to 80%.

Architected the backend of Buy Nothing, a social app serving 2M+ users, and delivered full scope on an investor-driven deadline. Built ETL pipeline for on-chain data from Solana, Ethereum, and Polygon for an NFT marketplace.

Improved application performance by 27% by optimizing resource loading and removing inefficient nested queries. Refactored an untested codebase into a tested system ahead of its move to microservices.

Built an ML-based smart waste-bin platform end to end in under six months with a small team, including Docker infrastructure and CI/CD.

Built the entire backend of a model-driven product that forecasts marketing spend, growing it from a one-file prototype into separate backend, frontend, and AI services.

Pursued a Bachelor of Engineering in Computer Science, completing the program in 2023.