Vladimir Turomsha

I’m a Python Backend Engineer with 3+ years building AI platforms, distributed systems, and microservices—focused on production reliability, performance, and clean architecture.

At Dolly AI (Dec 2024 – Present), I built a per-user fair-queue on Redis + Celery for a shared GPU pool, cutting tail wait times (15+ minutes down to under 1 minute). I also implemented GPU orchestration essentials like multi-provider AI integration with fallback/health tracking and a multi-agent coordination system (6 agents) that completed 126 tasks with zero critical regressions.

I care deeply about observability and operational safety. I deployed OpenTelemetry + Prometheus + Jaeger with a custom postgres exporter, reducing MTTR by ~40%, and I built a custom Celery transport for Azure Service Bus with priority routing and zero task loss.

Previously, I delivered internal automation and R&D work: RBAC document management with auditing, an InsightFace → pgvector face recognition pipeline with <50ms cosine search, and local LLM classification on Ollama (CPU-only) that improved auto-parse to 80%. I also developed real-time features on a sports platform (WebSocket + Redis pub/sub) and fixed race conditions in event flows using database-level safeguards.