Danylo Pilovets
telegram @nekyivskiy
HELLO
my name is
Danylo Pilovets

Platform & MLOps Tech Lead

building infrastructure since 2020

  • cloud / on-prem / hybrid
  • monitoring architectures
  • multi-cluster support
  • 0 production incidents
Consulting

Senior DevOps / SRE Engineer · Kyiv, Ukraine

About

I build the layer other engineers build on, and I own it end to end — from bare metal and cloud platforms up to the models serving live traffic.

Five years across five engineering organisations, on the same problem each time: turning an ambiguous infrastructure question into a platform other teams can use without asking me anything. Kubernetes orchestration, distributed data stores, high-throughput streaming, GPU compute, GitOps CI/CD and enterprise observability — built, scaled and carried in production. The work sits where Data Science, Software Development and Platform Reliability meet, which is usually where nobody wants to own it.

More than five exclusive contracts delivered so far: migrations across every weight class, systems rebuilt and hardened, and the audit and verification work that comes with both. At peak, more than twenty clusters carried at once, sitting under more than eight petabytes of data and an infrastructure bill in the millions of dollars — which is the part that makes cutting waste a finance decision rather than a tidiness exercise.

2020
Building infrastructure since
20+
Clusters run in parallel
8 PB
Data under management
5+
Exclusive contracts delivered
60%
Compute over-provisioning removed
0
Production outages caused

Consulting

ADMIT ONE$100/ hour

rate card

Consulting

Hourly, remote. A project gets a fixed scope first — this is the rate the clock starts at.

  1. 01Cloud, on-prem and hybrid

    The cluster and what it sits on.

    $100/ hour

  2. 02Multi-cluster Kubernetes

    The same estate, more than one cluster at once.

    $100/ hour

  3. 03GitOps and the developer platform

    So other teams ship without waiting on me.

    $100/ hour

  4. 04Observability

    Metrics, logs, alerts people actually read.

    $100/ hour

  5. 05AI / MLOps platform

    From an empty project to models in production.

    $100/ hour

  6. 06Cost and reliability

    Cut waste. Keep the thing up.

    $100/ hour
@nekyivskiyw.pilovets@gmail.com

Selected work

28apps live, from an empty project

A production AI platform, shipped solo in five months

AI Platform Architect / Lead MLOps · NDA (FinTech)

A one-year exclusive contract delivered in five months as the only engineer on it. An empty GCP project became a GKE platform running 28 GitOps-managed applications against live workloads, fronted by an internal developer platform that folds Argo CD, Prometheus and Alertmanager into one surface: cluster-wide health, deployment-to-alert correlation in real time, Alertmanager silencing with audit comments, and deep links straight into the right logs and dashboards. A plug-and-play config template lets teams onboard their own applications and clusters without going through me.

  • GKE
  • ArgoCD
  • Terraform
  • Helm
  • LiteLLM
  • LangGraph
  • Vertex AI
ISO 42001certified, end to end

An AI platform that survived a certification audit

AI Platform Architect / Lead MLOps · NDA (FinTech)

Took the GKE AI platform through complete ISO/IEC 42001 compliance, leading the architectural alignment with enterprise tech leads and enforcing AI management standards across cluster security, data governance and operational workflows. The security model underneath it is zero-trust by construction: HashiCorp Vault with KMS auto-unseal, External Secrets Operator, Entra ID SSO with fine-grained RBAC, VPN-only internal ingress, and scoped pod terminals issued to developers on demand.

  • Vault
  • External Secrets
  • Entra ID SSO
  • OPA
  • GitLab CI/CD
60%of over-provisioning, removed

Cost treated as a reliability metric

MLOps Engineer · Playtika · SRE Tech Lead · Oxtech

Cut Kubernetes request over-provisioning by roughly 60% across three production clusters, on estates whose infrastructure bill ran into the millions of dollars — at that size a percentage is a budget line, not a housekeeping metric. The same discipline applied earlier took another cluster estate down by 40%, and migrating projects onto hardware already rented from the vendor saved a further 20%.

  • Kubernetes
  • Helm
  • Prometheus
  • Grafana
  • RunAI
80%fewer Kubernetes alerts

Observability that people actually read

SRE Tech Lead · Oxtech · DevOps Engineer · Wix

Designed monitoring architectures that pushed teams toward observability-driven development and cut Kubernetes alerts by 80%, making the system legible to engineers new to Kubernetes rather than just quieter. Shipped later as Observability-as-Code — Prometheus, Grafana, Loki and Alloy with a custom three-tier noise-reducing ruleset and automated Slack routing. On the way, terabytes of Victoria Metrics TSDB were migrated onto new environments with no downtime, no data loss and no target loss.

  • Prometheus
  • Victoria Metrics
  • Grafana
  • Loki
  • Alloy
  • Ansible

Architecture

Abstracted from a system in production: names removed, structure intact. Five of the seven bands come out the same whoever I build them for.

Core — Kubernetes, on whatever it happens to sit on

A namespace per service · a throwaway namespace per merge request · identity over static keys

01Substrate — swappable

Managed cloud cluster

Node pools, cloud KMS and platform-issued identity, with the provider doing the parts it does better than I would.

Self-managed on bare metal

The cluster, the operating system underneath it and the configuration management that keeps the fleet identical, all owned outright.

On-prem virtualisation

Clusters on hypervisors somebody else's capacity plan already paid for, with lifecycle management built on top.

Hybrid estate

Part in a cloud, part on owned hardware, and workloads moved between the two without the applications noticing.

Network

Private networks, routing, load balancers and VPN — including deciding what is allowed to be reachable at all.

Storage

Volumes, distributed filesystems, and the rotation and capacity work that stops a disk from becoming an incident.

02Developer surface — invariant

Internal developer platform

One screen for every team: cluster health, deployment correlated with the alert it caused, alerts silenced in place with an audit comment, and deep links into the right dashboard. Teams onboard their own applications through a template.

03Control plane — invariant

GitOps controller

A root application renders the whole catalogue. A merge request gets its own namespace.

Secret store + operator

Auto-unsealed by a key management service, a role and policy per service, delivered into the pod and rotated by a reloader.

Ingress

A single entry point, internal by default, certificates from a CA that lives in the cluster.

04Deployment unit — invariant

Umbrella chart

Several subcharts released, versioned and rolled back as one thing. Composition is vertical.

Standalone chart

One workload on its own version and its own cadence, for anything with an independent lifecycle.

Horizontally scaled workload

No shared volume, so replicas are added and removed freely.

Stateful workload

Operator-managed: a primary with replicas, spread by anti-affinity.

05Data — invariant

Relational store

Run by an operator. A new database or role is two lines of values; metrics land in the shared collector.

Queue

Backs asynchronous execution so workers stay stateless.

Shared rate limiting

One budget for model traffic across every application.

Certificates

Internal authority, wildcard across cluster names.

06Payload — swappable

Services under live traffic

Product backends and microservices, carried on call, with the reliability work that implies.

Streaming and queues

High-throughput pipes where back-pressure and retention are the design, not a setting.

Batch and scheduled work

Orchestrated data pipelines, where a failure is noticed hours later unless the platform says so first.

Databases for other teams

Relational and NoSQL stores run as a service: replication, integration and not losing data across failure modes.

High-load backends

Systems sized for peaks rather than averages, where over-provisioning is the easy answer and the wrong one.

Model serving

One in-cluster gateway for every model call, so provider keys never enter an application, with embeddings and document parsing running locally.

07Observability — invariant

Metrics

Own rules across three severity tiers, written as code.

Alert routing

Severity decides the channel and whether anyone is woken up.

Dashboards as code

Versioned in the same repository as everything they describe.

Log pipeline

An agent on every node ships pod logs, cluster events and stack traces to be queried next to the metrics that explain them.

Every decision here is written down beside the code that implements it, together with the alternative that was rejected — and CI fails the build if a documentation link stops resolving to a real line.

Approach

If it isn't in git, it doesn't exist

28 applications managed through Argo CD, cluster-wide deployments automated with FluxCD, and a backlog framework that is itself a git repository — 500+ tasks executed deterministically with zero context drift.

Observability is a deliverable

Monitoring shipped as code with a three-tier ruleset and automated routing, custom exporters written where the platform had no visibility, and post-incident reviews aimed at the systemic cause rather than the symptom.

Secrets never reach application code

Vault with KMS auto-unseal, External Secrets Operator, and an in-cluster LLM gateway on Workload Identity that removed raw API keys from applications entirely.

A platform is a product; teams are its users

Self-service onboarding templates, eight shared CI/CD pipeline templates with OPA policy enforcement and Helm schema validation, and a directory that routes each team to its own Vault, dashboards and tooling.

Take the problem nobody wants

Full ownership of complex, highly ambiguous engineering challenges — including the ones others avoided — driving root cause analysis through to a long-term infrastructure answer.

Migrations are judged on what didn't happen

Legacy TSDB clusters deprecated and replaced without downtime, data loss or target loss; monitoring moved off Chef onto Ansible; CI runners moved off bare metal into Kubernetes pods. Across five years of this, on estates of twenty-plus clusters, no production outage has started with me.

Stack

Orchestration & Compute

  • Kubernetes
  • Helm
  • Docker
  • Rancher
  • Linux
  • GPU workloads
  • vLLM
  • Nginx

Cloud & IaC

  • AWS
  • GCP
  • Terraform
  • Terragrunt
  • Ansible
  • SaltStack
  • Chef
  • Cloudstack
  • Vault

GitOps & CI/CD

  • ArgoCD
  • FluxCD
  • GitLab CI/CD
  • Jenkins
  • Git
  • OPA policy

Observability

  • Prometheus
  • Victoria Metrics
  • Grafana
  • Loki
  • Alloy
  • Alertmanager

ML & AI Platform

  • Kubeflow
  • Vertex AI
  • RunAI
  • LiteLLM
  • LangGraph
  • Langfuse
  • Arize Phoenix
  • DeepEval
  • MCP
  • Airflow

Data stores & Streaming

  • PostgreSQL
  • MySQL
  • MongoDB
  • Aerospike
  • Redis
  • Kafka
  • RabbitMQ
  • Hadoop
  • Qdrant
  • pgvector

Languages

  • Python
  • Go
  • Bash
  • Java
  • TypeScript
  • SQL/NoSQL

Experience

AI Platform Architect / Lead MLOps · NDA (FinTech)

June 2026 — Present

  • Delivered a one-year exclusive contract in five months as the only engineer on the project: designed and shipped a production AI platform on GKE from an empty GCP project to 28 GitOps-managed applications serving live workloads.
  • Single-handedly architected and built an “AI-Enabled Internal Developer Platform”. Unified Argo CD, Prometheus, and Alertmanager into a multi-team IDP featuring cluster-wide health metrics, real-time deployment-to-alert correlation, and direct deep-links to logs and dashboards, plus in-place Alertmanager silencing with audit comments and an embedded business AI assistant.
  • Designed a plug-and-play configuration template enabling teams to self-onboard their applications and clusters, providing a centralized ecosystem directory with auto-routed links to team-specific Vaults, Grafana dashboards, and internal tooling.
  • Engineered agentic execution workflows using LangGraph and consolidated LLM access via an in-cluster LiteLLM gateway with Vertex AI (Workload Identity) and Redis-backed rate limiting, removing raw API keys from applications.
  • Authored and published tasks-as-code — a git-native backlog framework enabling deterministic task execution and automated context alignment for human and AI-agent collaboration across 500+ tasks with zero context drift.
  • Deployed Observability-as-Code (Prometheus, Grafana, Loki, Alloy) featuring a custom 3-tier noise-reducing alerting ruleset and automated Slack routing.
  • Enforced zero-trust security using HashiCorp Vault (KMS auto-unseal), External Secrets Operator, Entra ID SSO with fine-grained RBAC, and internal VPN-only ingress alongside dynamic scoped pod terminals for developers.
  • Engineered reusable GitLab CI/CD pipeline templates — 8 shared templates with automated OPA policy enforcement, Helm schema validation, and doc-link integrity checks — to standardize build, test, and release cycles across all engineering teams.
  • Navigated the GKE AI platform through complete ISO/IEC 42001 certification compliance, leading architectural alignments with enterprise tech leads and enforcing strict AI management standards across cluster security, data governance, and operational workflows.
  • Kubernetes
  • GCP
  • Vertex AI
  • Terraform
  • Helm
  • ArgoCD
  • LiteLLM
  • LangGraph
  • Langfuse
  • pgvector
  • Docling
  • GitLab CI/CD
  • Vault
  • Entra ID SSO
  • PostgreSQL
  • Redis
  • Kafka
  • Qdrant
  • Prometheus
  • Grafana
  • Loki
  • Alloy
  • TypeScript
  • Python
  • Bash

DevOps Engineer · Wix

Kyiv, Ukraine · December 2025 — June 2026

  • Managed and supported large-scale Kubernetes environments running business-critical workloads.
  • Led migration of monitoring infrastructure from Chef to Ansible, with standardization and automation.
  • Performed seamless migrations of large Victoria Metrics TSDB instances, deprecating legacy storage clusters and introducing new environments without downtime, data loss, or target loss.
  • Designed and implemented new monitoring environments, data sources, Grafana dashboards, and observability workflows.
  • Participated in incident response, troubleshooting, and RCA.
  • Automated monitoring infrastructure provisioning and configuration using Terragrunt and Terraform.
  • Developed custom exporters and metrics to improve visibility into platform and application health.
  • Designed, developed, and maintained Helm charts for monitoring and observability services deployed in Kubernetes.
  • Kubernetes
  • Helm
  • Grafana
  • Victoria Metrics
  • AWS
  • Python
  • Bash
  • Chef
  • Ansible
  • Java
  • Terraform

MLOps Engineer · Playtika

Kyiv, Ukraine · May 2023 — April 2026

  • Supported machine learning workflows on Kubernetes with GPU-based workloads in production and training environments.
  • Designed and maintained custom Helm charts for machine learning infrastructure and services.
  • Implemented a robust log and data rotation strategy across numerous services, preserving terabytes of free disk space and preventing downtime from storage issues.
  • Built custom alerting and monitoring systems tailored to the needs of ML and data services.
  • Worked with a wide range of NoSQL and SQL databases, ensuring efficient storage, replication, and integration with distributed systems while preventing data loss across failure scenarios.
  • Ensured 24/7 availability and reliability of Kubernetes clusters supporting critical production workloads, and participated in on-call rotations.
  • Led post-incident reviews to uncover systemic issues, improve observability, and prevent recurrence of critical failures.
  • Optimized Kubernetes resource usage by reducing request over-provisioning by approximately 60% across three clusters, significantly improving efficiency and cutting operational cost.
  • Migrated multiple operators and applications from a kubespray setup to the Palette platform via Cloudstack, delivering more consistent deployments and streamlined lifecycle management.
  • Kubernetes
  • Helm
  • Aerospike
  • Hadoop
  • MySQL
  • Kafka
  • Airflow
  • Kubeflow
  • RunAI
  • Nginx
  • Prometheus
  • Grafana
  • ArgoCD
  • FluxCD
  • Jenkins
  • Ansible
  • Cloudstack
  • MCP
  • vLLM
  • Arize Phoenix

SRE Tech Lead · Oxtech

Tel Aviv, Israel · April 2021 — May 2023

  • Managed and maintained bare-metal Kubernetes (including Rancher) and debugged Linux-based applications.
  • Migrated projects between Kubernetes clusters, resulting in 20% cost savings by leveraging existing vendor-rented hardware.
  • Created and maintained custom Helm charts.
  • Developed and implemented monitoring architectures, fostering observability-driven development, reducing Kubernetes alerts by 80%, and making the system more approachable for Kubernetes newcomers.
  • Created and maintained GitLab CI/CD pipelines for continuous integration and deployment.
  • Supported backend developers and SRE colleagues with infrastructure automation, CI/CD pipelines, and monitoring tools.
  • Automated cluster-wide infrastructure deployments using FluxCD, ensuring GitOps-driven consistency and speeding up environment provisioning.
  • Optimized Kubernetes resource usage by reducing request over-provisioning by 40%, lowering cluster load and improving cost efficiency.
  • Migrated CI/CD execution from bare-metal GitLab Runners to Kubernetes-based runner pods, increasing scalability and fault tolerance.
  • Kubernetes
  • SaltStack
  • Docker
  • GitLab CI/CD
  • Redis
  • MongoDB
  • RabbitMQ
  • Vault
  • Helm
  • Linux
  • Bash
  • Prometheus
  • Grafana
  • Loki
  • FluxCD
  • ArgoCD

DevOps Engineer · Atlassian

Remote, Ukraine · January 2020 — April 2021

  • Worked with cloud infrastructure (primarily AWS, with partial GCP usage) to support internal platforms and services.
  • Supported and improved internal CI/CD pipelines, ensuring stable and predictable delivery workflows across multiple teams.
  • Collaborated with engineering teams to maintain and troubleshoot Kubernetes-based services, focusing on reliability and performance.
  • Contributed to infrastructure optimization initiatives, including resource tuning and cost visibility improvements.
  • Participated in maintaining workflow orchestration (Airflow) within containerized environments.
  • Kubernetes
  • Terraform
  • Docker
  • Helm
  • Ansible
  • Airflow
  • AWS
  • GCP
  • ArgoCD
  • Grafana
  • Bash
  • Python

Education

M.S., Central Ukrainian State University · 2018 — 2024

Contact

Download CV →EmailTelegramLinkedIn