At TERAMIND, I reduced LLM token consumption by 60% by replacing high-volume AI agent interactions with direct API integrations while maintaining comparable output quality. I also developed a BERT-based URL categorization model for isolated customer environments, enabling fully offline inference.
At Grid Dynamics, I scaled IoT Hub support from 5K to 30K concurrent device connections by optimizing database performance and migrating to asyncio with FastAPI. For CRITICALSTART, I improved a Celery pipeline handling 5–7M security events daily and reduced ElasticSearch load by 66%, saving $75K in annual AWS costs.
Earlier, at Deloitte, I streamlined a Celery processing pipeline and helped reduce translation costs by 40%. At Bmetric, I implemented a Django REST Framework API endpoint for generating configuration files used by 4M+ clients daily.
