Skip to content

Resume

Vitalii Serbyn

AI Platform / MLOps Engineer

serbyn.vitalii@gmail.comKyiv, Ukraine · Remote only · UK LTD for contractingserbyn.pro

Professional Summary

Senior engineer with 12+ years shipping production systems in mobile, web, and cloud. I design and run reliable LLM services with clear SLOs, CI/CD, MLflow registries, cost controls, and real observability. I have led small teams, owned release trains, and I bring that discipline to building and operating AI platforms that are easy to deploy, measure, and roll back.

Technical Skills

LLM & GenAI (Current Focus)

LangChain/LangGraph
vLLM
OpenAI API
Ollama
BLIP-2
RAG
Hybrid Retrieval

MLOps & Platform (Learning)

MLflow
SLO-gated CI
A/B Evaluation
Experiment Tracking
Model Registry
Canary Deployment

Backend & Data (12+ years)

Python
FastAPI
Node.js/NestJS
Celery
PostgreSQL
Redis
RabbitMQ
Qdrant

Cloud & Infrastructure (Production)

Kubernetes
Docker
AWS ECS/EKS
GCP/GKE
GitHub Actions
Terraform basics

Observability & Ops (Expert)

Prometheus
Grafana
Jaeger
Sentry
CloudWatch
PagerDuty

Frontend & Mobile

Flutter
Android/Kotlin
React basics

Professional Experience

AI Platform / MLOps Engineer

Easelect LTD • Remote

2023 - Present
  • Built Threads-Agent GenAI platform with MLflow registry, SLO-gated CI, and production observability
  • Implemented ROI-Agent for automated media buying with multimodal pipeline (BLIP-2 + LLM analysis)
  • Created Achievement Collector for PR impact analysis and automated portfolio generation
  • Developed reliable LLM services with clear SLOs, cost controls, and real observability
  • Applied production discipline to AI platforms: explicit SLOs, fast rollback, cost budgets

Senior/Lead Engineer

EPAM, GlobalLogic, Startups • Remote/Hybrid

2013 - 2023
  • Led teams of 3–4 engineers, owned release trains with 99.5+ uptime for mobile apps (5M+ MAU)
  • Optimized backend APIs serving 1M+ requests/day with p95 latency <200ms (Redis caching, query optimization)
  • Implemented A/B testing frameworks processing 100k+ events/day for user behavior analysis
  • Built event-driven architectures with message queues handling 50k+ messages/min peak load
  • Established monitoring with Prometheus/Grafana: 15+ custom metrics, alert rules, SLA tracking
  • Reduced deployment time from 45min to 3min using Docker + Kubernetes + automated rollback

AI Relevance: These same patterns apply to LLM services: p95 latency SLOs, request/response caching, event processing for model inputs, monitoring for drift detection, and fast rollback for model deployments.

Key Projects

Threads-Agent — GenAI Content Platform

LangGraph • MLflow • Kubernetes • Prometheus • 2024 • Personal R&D

  • Multi-service system: persona runtime, optimizer service, publishing adapters
  • MLflow model registry with evaluation jobs tied to CI for automated quality gates
  • SLO-gated deployments on p95 latency, error rate, and token-cost deltas
  • Full observability stack: Prometheus metrics, Grafana dashboards, Jaeger tracing

ROI-Agent — Media Buyer Automation

BLIP-2 • Ollama • Meta Ads API • Celery • 2024 • Personal R&D

  • Multimodal pipeline: BLIP-2 for creative captions, LLM analysis for copy variants
  • Hybrid inference: local Ollama (Mixtral) with OpenAI fallback and structured output
  • A/B testing with sequential decision rules, performance rollups, policy guardrails
  • Integrations: Facebook Graph API, Shopify webhooks, Slack to Linear automation

Achievement Collector — Portfolio Automation

GitHub API • Impact Analysis • Portfolio Generation • 2024 • Personal R&D

  • GitHub PR analysis extracting technical metrics and generating ImpactScores
  • Automated portfolio generation from development activity and MLflow experiments
  • CI integration for consistent reporting and resume bullet generation

High-Traffic Mobile Backend Infrastructure

Node.js • PostgreSQL • Redis • Kubernetes • 2018-2022 • EPAM/GlobalLogic

  • API gateway handling 1M+ requests/day with circuit breakers and rate limiting
  • Real-time event processing: 100k+ events/day through Redis Streams and message queues
  • Database optimization: query performance tuning reducing p95 response time from 800ms to 120ms
  • Monitoring implementation: 20+ custom Prometheus metrics with Grafana dashboards and PagerDuty alerts

Education

M.Sc. in Computer Science

V. N. Karazin Kharkiv National University

2009 - 2011

Software Engineering Management

EPAM School

2019

Business Setup & Legal

Contracting

  • UK LTD available for contracting
  • Comfortable with W-8BEN-E for US clients
  • Standard business terms available

Payments

  • Wise Business or Revolut Business
  • Crypto: USDC/USDT (ERC-20, SOL, TRX)
  • Invoices in USD, EUR, or GBP
  • Standard contracting terms available

Languages

English: Fluent
Ukrainian: Native
Russian: Native