Konstantinos Kosyfarinis

Backend Engineer — Platform, DevOps & SRE

Backend engineer focused on platform engineering, DevOps, and SRE. I design and ship the infrastructure layer — APIs, distributed agents, Kubernetes clusters, GitOps pipelines, and the observability that keeps petabyte-scale data stores honest, on bare metal and in the cloud.

See the work GitHub

Remote-friendly · usually online, occasionally asleep

kosy@space:~

Platform EngineeringDevOpsSite ReliabilityKubernetesGitOpsBackend SystemsObservability

Infrastructure at scale

Some numbers from systems I've built, migrated, or kept running.

50B+

events in ClickHouse

Operated a ClickHouse cluster ingesting more than 50 billion events, tuned for sub-second queries at that scale.

15TB+

MongoDB in production

Ran and tuned MongoDB deployments north of 15TB — sharding, indexing strategy, and operational recovery included.

downtime on a 5TB MySQL cut-over

Migrated a 5TB MySQL production primary to new hardware with zero seconds of downtime.

1000+

cores across a Proxmox cluster

Built and run a Proxmox cluster spanning 1000+ CPU cores, 5TB of RAM, and 60TB+ of raw storage.

By hand

Kubernetes, the hard way

Deployed every control-plane and node binary manually, following Kelsey Hightower's Kubernetes the Hard Way — no kubeadm, no managed control plane.

GitOps

ArgoCD, Rollouts & workload identity

Progressive delivery via Argo Rollouts, reconciliation via ArgoCD, Keycloak SSO, and GCP workload identity bound straight to Kubernetes service accounts.

Selected work

Three platforms, built end to end — schema to shell to deploy.

Zephra

In production

AWX / Semaphore competitor

An Ansible execution platform built from the ground up.

A full Ansible automation platform: connect git repos, define playbooks/inventories/credentials, and run them on distributed Go agents that pull work over a poll loop rather than the backend ever touching a shell. Project-scoped RBAC, HMAC-signed outbound webhooks with retry + delivery logs, multi-vault-ID Ansible runs, live SSE log streaming, and a global search layer all sit on top of that core loop.

Sole engineer — backend API, agent daemon, dashboard, and deploy pipeline.

Agents poll for work and execute Ansible locally — the control plane never runs a subprocess
Viewer / operator / admin RBAC enforced per project, closing an org-membership IDOR
Selectable retry — full rerun or failed-hosts-only — never automatic
Reliable notifications: HMAC-SHA256 signed payloads, exponential-backoff retries, full delivery log

NestJSMikroORMPostgreSQLGoNext.jsReact QueryDockerNginxAnsible

Saryno Observability

In production

OpenTelemetry ingest platform

An OTLP-native tracing backend built for high-cardinality ingest.

A traces ingestion pipeline that speaks OpenTelemetry natively: a Go agent runs local OTLP HTTP/gRPC servers, batches and gzips spans, and forwards them to a NestJS ingest API that denormalizes hot fields (HTTP method, status, route) before writing into ClickHouse for sub-second query at scale. Public read-only CMS API fronted by nginx-level path filtering for the blog/content layer.

Backend + agent engineer — ingest API, OTLP parsing pipeline, ClickHouse schema.

Local OTLP HTTP (4318) + gRPC (4317) servers batch spans (1000/1s) before shipping upstream
Backend denormalizes attributes into hot columns + materialized views for fast trace search
Scoped API keys (ingest:spans) with Bearer/X-API-Key auth and gzip-aware ingestion
Public CMS API locked to GET /api/articles at the nginx layer — everything else 403s

NestJSGoClickHouseOpenTelemetry (OTLP)gRPCStrapiNginx

TaskTown

Shipped

Multi-vertical marketplace

A real-estate / product / service marketplace with live buyer-seller messaging.

A marketplace spanning three listing verticals — real estate, products, and services — with a seller manager dashboard and a public buyer-facing storefront. Built geo-tagged real-estate search (radius + bounding filters), a full messaging system with read receipts, and shipped in English, Greek, and German from day one.

Backend + full-stack engineer — API, manager dashboard, and public marketplace.

Geo radius search on real-estate listings (lat/lon + km radius) alongside price/bed/bath filters
Buyer ↔ seller messaging with read receipts and unread counts, polled in near-real-time
Fully localized UI (en/el/de) across both the manager and public marketplace apps
Map-based location picker (Leaflet + Nominatim) wired into listing creation

NestJSMikroORMPostgreSQLNext.jsLeafleti18next

Stack

What I reach for, roughly in order of how often it shows up.

Languages

TypeScriptGoSQLBash

Backend & data

NestJSMikroORMPostgreSQLMongoDBMySQLRedisBullMQClickHouse

Kubernetes & GitOps

KubernetesArgoCDArgo RolloutsKeycloakGitOps

Cloud & IaC

GCPPub/SubTerraformOpenTofuHashiCorp VaultCrossplane

Infra & virtualization

ProxmoxBare metalDockerNginxAnsibleCloudflareLinuxsystemd

Observability

OpenTelemetryOTLPStructured loggingSSE

Frontend (when needed)

Next.jsReactTailwind CSSReact Query

How I work

A few things that stay true across every project above.

End to end, on purpose

I design the schema, build the API, write the agent or worker that acts on it, and deploy the whole thing — so nothing gets lost in a handoff.

Boring infrastructure, on purpose

Distributed systems fail in the corners. I'd rather ship a poll loop that's obviously correct than a push architecture that's clever and fragile.

Make the failure visible

Retries are explicit, not silent. Deliveries are logged, not fire-and-forgotten. If something can fail, the operator should be able to see it happen.

Let's build something that stays up

If you need someone who's comfortable owning the whole path from schema to production, find me on GitHub.

github.com/Kosyness