Developed in-house ML model quality monitoring pipeline (Airflow DAGs with Spark integration hosted on GCP).Developed SQL schema & data generation from natural language prompt service (asynchronous gRPC server with Langchain integration) -- https://www.synthesized.io/post/leveraging-ai-for-diverse-test-data-generation-in-software-testing

Contributed to both backend (web app, async CPU-heavy task scheduler, other microservices) and ML submodules (yield productivity maps, yield data analysis, soil sampling).Designed and implemented various other features (billing & monetization, monitoring, data anomaly detection etc.).Achievements:Collaborated with an agronomist to develop and implemented harvester’s yield data analysis framework — ~100k users on over a million fields worldwide in 2022 — https://blog.onesoil.ai/en/how-to-conduct-field-tests Developed and implemented soil sampling method based on segmentation of the productivity map — main selling point for LATAM-based farming companies — https://blog.onesoil.ai/en/why-is-conducting-soil-nutrient-analysis-optional Refactored parts of existing Django application to reduce costs (~43% on AWS) and improve performance (~20% speed-up in data-heavy pipelines) while maintaining 99.9% uptime.

Developed Python & JVM-powered backend for SaaS cloud platform (GCP-hosted) providing ML model inference through a public API.Achievements:Implemented the Python web application project end-to-end: system design, extensive multi-level testing, documentation, deployment and monitoring.Led multiple successful pilots (Deutsche Bank, UBS, UK W&P) -- https://www.synthesized.io/post/will-your-data-pass-the-test

Developed web app service with on-premise deployed data collection agent.Led a team of 3 interns to build a backend for medical data collection & processing.Achievements:Managed integration with several municipal hospitals (demo → pilot → contract)

Designed low-shot image recognition and joint text+image learning pipelines.Developed a special case of the image stitching problem.Led the project on automatic shelf monitoring with X5.ru.