NowBuilding Forbes Vetted backends

Vahan SahakyanSenior Backend Engineer

5+ years designing distributed systems end to end, from the data model and APIs to event streaming and the Kubernetes clusters they run on.

backends you stop worrying about

Shipped for

  • Forbes
  • Warner Music Group
  • Optimum Partners
  • Datafoundry

01 About

Senior backend engineer.Data model to cluster.

Senior backend engineer with 5+ years designing and building distributed systems, from the data model and API design to the Kubernetes clusters they run on.

years building backends
5+
Go services in the DSN reference system
11
code coverage kept while shipping features
100%
companies shipped for
4
languages
Go, Python, TypeScript, Java
streaming
Kafka, Redpanda, Pub/Sub
data
PostgreSQL, ScyllaDB, ClickHouse, MongoDB
infra
AWS, GCP, Kubernetes, ArgoCD
observability
Datadog, Prometheus, Grafana

02 Approach

One write, end to end.

One post through a system I built. Each service owns its data, and events, not shared tables, carry state between them.

  1. The gateway checks the JWT against Keycloak's JWKS and turns HTTP/JSON into a gRPC call with a 5s deadline.

  2. posts-service writes the post and its outbox event in one ScyllaDB LOGGED BATCH. Both exist or neither does, so there is no dual write to reconcile.

  3. A relay reads the outbox and publishes post.created to Redpanda. Delivery is at-least-once, so consumers are built for redelivery.

  4. Feed, notification, search and the event writer each consume at their own pace. Feed and notification writes are idempotent, and the event store collapses duplicate ids.

  5. A poison message retries 3 times with backoff, then moves to the dead-letter topic. Alerts fire on DLQ depth, outbox backlog and consumer lag.

  6. Every event lands in ClickHouse, 500 messages or 200ms per insert. Feeds in Valkey are a read model: the rebuilder replays the store into a fresh copy and renames it in.

dsn / one post, end to end

ClientGatewayPostsScyllaDBRedpandapost.dlqFeedNotifySearchEvent writerValkeyRebuilderClickHouseClientGatewayPostsScyllaDBRedpandapost.dlqFeedNotifySearchWriterValkeyRebuilderClickHouse
RequestEventDataDead letter

03 Work

Case studies,drawn as the systems they are.

Problem, architecture, result. Each diagram is the real service topology: filter by edge type, trace a node's dependencies, zoom in.

Distributed Social Network

11 Go services over gRPC and Redpanda, with a transactional outbox, a CQRS feed and a ClickHouse event store. OpenTelemetry, Tempo, Loki and Prometheus for observability; deployed to Kubernetes with Helm and Argo CD.

  • Go
  • Microservices
  • Kafka
  • CQRS
  • PostgreSQL
  • ClickHouse
  • Kubernetes
01 Problem
Posts, comments and likes have to reach feeds, notifications and search without a shared database or distributed transactions, and survive broker redelivery and poison messages.
02 Architecture
Each service owns its store. Writes commit the row and an outbox event together; a relay publishes to Redpanda, where feed, notification, search and event-writer consumers fan out independently.
03 Result
  • No dual writes: an event is published only if its row committed
  • Idempotent consumers, so redelivery is safe
  • Poison messages retry 3 times with backoff, then dead-letter
  • Feeds are a CQRS read model, rebuildable from the event store
GatewayKeycloakArgo CDUsersPostsCommentsLikesMediaRebuilderRedpandaTelemetryFeedNotifySearchEvent writerScyllaDBPostgreSQLMinIOValkeyClickHouseElasticGatewayKeycloakArgo CDUsersPostsCommentsLikesMediaRebuilderRedpandaTelemetryFeedNotifySearchEvent writerScyllaDBPostgreSQLMinIOValkeyClickHouseElasticGatewayKeycloakArgo CDUsersPostsCommentsLikesMediaRebuilderRedpandaTelemetryFeedNotifySearchEvent writerScyllaDBPostgreSQLMinIOValkeyClickHouseElastic

11 Go services, one event backbone

Hover a node to trace its edges. Drag to pan, Ctrl/Cmd + scroll to zoom.Tap a node to trace its edges. Pinch with two fingers to zoom.

  • Edge
  • Go service
  • Streaming
  • Data store
  • Platform

Event stream simulated: offsets, lag, retries, DLQ

Code Inspect

Code review platform: assignments are submitted as GitHub branches and tracked through review states. Spring Boot API on MySQL; changes stream through Kafka to a WebSocket server for live updates. Runs on Kubernetes behind an NGINX Ingress.

  • Java
  • Spring Boot
  • MySQL
  • Kafka
  • WebSockets
  • Kubernetes
01 Problem
Students and reviewers work on the same assignments and need to see review state change as it happens, across three roles.
02 Architecture
A Spring Boot API on MySQL owns the 6-state review flow and publishes every change to a Kafka topic; a stateless Node.js WebSocket server consumes it and pushes to open dashboards.
03 Result
  • Live dashboards without polling the API
  • Stateless WebSocket tier, separate from the API
  • JWT roles enforced in one Spring Security filter
  • Same stack runs on Docker Compose or minikube
BrowserminikubeIngressFrontendBackendWS serverKafkaZooKeeperMySQLBrowserminikubeIngressFrontendBackendWS serverKafkaZooKeeperMySQLBrowserminikubeIngressFrontendBackendWS serverKafkaZooKeeperMySQL

Spring Boot API, live updates over Kafka and WebSockets

Hover a node to trace its edges. Drag to pan, Ctrl/Cmd + scroll to zoom.Tap a node to trace its edges. Pinch with two fingers to zoom.

  • Edge
  • Service
  • Streaming
  • Data store
  • Platform

More on GitHub

04 Capabilities

Grouped bythe problems they solve.

Seven areas, each with the tools used in production. The map below links tools used on the same project.

  • 01

    Distributed Systems

    Microservices, event-driven patterns (Kafka/Redpanda), CQRS and event sourcing, in production and in an open-source POC with 10+ services.

    • Microservices
    • CQRS
    • Event sourcing
    • Transactional outbox
    • gRPC
  • 02

    Event Streaming

    Kafka-compatible brokers with async consumers: idempotent handlers, retries with backoff, dead-letter queues, and alerts on consumer lag.

    • Kafka
    • Redpanda
    • Pub/Sub
  • 03

    Cloud & Kubernetes

    Multi-tenant prod/staging/dev isolation, GitOps provisioning with ArgoCD and Argo Events, Gateway API.

    • Kubernetes
    • GCP
    • AWS
    • ArgoCD
    • Argo Events
    • Helm
  • 04

    Data

    MongoDB at Forbes, PostgreSQL, ScyllaDB for time-series, ClickHouse for event stores. Chosen per workload: relational, document, time-series or analytical.

    • MongoDB
    • PostgreSQL
    • ScyllaDB
    • ClickHouse
    • Valkey
  • 05

    Languages

    Go for performance-critical services, Node.js/TypeScript for full-stack work, Java/Spring in previous roles. REST and gRPC APIs in production.

    • Go
    • TypeScript
    • Node.js
    • Python
    • Java
  • 06

    Observability

    Metrics, distributed tracing and structured logging as core requirements. Prometheus, Grafana, Jaeger, Loki.

    • Prometheus
    • Grafana
    • Jaeger
    • Loki
    • Datadog
    • OpenTelemetry
  • 07

    Leadership

    Own architectural decisions from data model to deployment. Design docs, code review, mentoring.

    • Design docs
    • Code review
    • Mentoring

Stack, linked by project

GoTypeScriptJavagRPCRESTKafkaRedpandaPostgreSQLMongoDBScyllaDBClickHouseMySQLValkeyElasticsearchKubernetesDockerArgoCDArgo EventsGCPAWSPrometheusGrafanaLokiOpenTelemetryGoTypeScriptJavagRPCRESTKafkaRedpandaPostgreSQLMongoDBScyllaDBClickHouseMySQLValkeyElasticsearchKubernetesDockerArgoCDArgo EventsGCPAWSPrometheusGrafanaLokiOpenTelemetry

Hover or tap a technology to see where it was used.

End to end

From the data modelto the cluster it runs on.

I take a service the whole way: model the data, build the API and the events around it, then ship it to Kubernetes through GitOps and watch it in production.

git push && argocd app sync

Data modelServiceCluster

05 Experience

A career,read as a git log.

Employer on one lane, client work on the other. Select a commit to jump to the role.

  1. Senior Backend Engineer

    Forbes · Contract

    May 2024 - Present · 2 yrs 7 mosJersey City, New Jersey, United States · Remote

    Platform Team, Bertie CMS, Vetted Team

    • Contributing to multi-tenant Kubernetes infrastructure across prod/staging/dev with namespace isolation
    • Backend development on event-driven systems (Kafka-compatible brokers, async patterns)
    • Working with GitOps automation (ArgoCD, Argo Events) for infrastructure provisioning
    • Production MongoDB experience: schema design, scaling concerns, consistency patterns
    • Code reviews and mentoring junior engineers on backend/systems patterns
    • Full-stack contributions across frontend, backend, and infrastructure layers
    • Go
    • Node.js
    • TypeScript
    • Kubernetes
    • Docker
    • MongoDB
    • PostgreSQL
    • ArgoCD
    • Argo Events
    • Prometheus
    • Grafana
    • GCP
    • Linux
    • REST APIs
    • React
  2. Senior Backend Engineer

    Optimum Partners - Armenia · Full-time

    May 2024 - Present · 2 yrs 7 mosYerevan, Armenia · Hybrid

    Embedded at Forbes: Platform & Vetted Teams

    Senior Backend Engineer · Jan 2025 - Present

    • Architecting distributed Go and TypeScript services for Forbes Vetted: search, cursor pagination, scraping, data pipelines
    • Event-driven architecture on Kafka-compatible brokers with async consumers
    • Multi-tenant Kubernetes on GCP with namespace isolation and GitOps (ArgoCD, Argo Events)
    • Leading backend delivery on Nexus initiatives; mentoring engineers on system design

    Full Stack Software Engineer · May 2024 - Jan 2025

    • Go and TypeScript services and integrations for Bertie CMS and Vetted
    • Integrated the Zephr and Cordial platforms
    • Maintaining 100% code coverage while shipping new features
    • Technical documentation: architecture decisions, runbooks, design patterns
    • Go
    • Node.js
    • TypeScript
    • Kubernetes
    • Docker
    • Microservices
    • MongoDB
    • PostgreSQL
    • Kafka
    • ArgoCD
    • GCP
    • Linux
    • Technical Leadership
  3. Software Engineer (Backend & Infrastructure)

    Datafoundry · Full-time

    Sep 2023 - May 2024 · 9 mosYerevan, Armenia · Hybrid

    DF mSignal AI, DF Literature Monitor, DF mSafety AI

    • Led large-scale refactoring across microservices architecture
    • Set up and maintained frontend quality tools (ESLint) in legacy codebases
    • Achieved minimum 70% unit test coverage, reduced code smells for SonarQube quality gates
    • Restructured static components into reusable library-style modules for ecosystem-wide use
    • Technical leadership and peer mentoring on code quality and design patterns
    • JavaScript
    • TypeScript
    • React
    • Spring
    • Node.js
    • Code Review
    • Technical Design
    • ESLint
    • Webpack
    • Jest
    • Technical Leadership
  4. Software Engineer

    Warner Music Group · Contract

    Nov 2022 - Sep 2023 · 11 mosNew York, United States · Remote

    ADA (Alternative Distribution Alliance) & DSPs (Spotify, Apple Music, etc.)

    Music Royalty and Publishing Platform

    • TypeScript
    • React
    • Design Patterns
    • Java
    • Business Analysis
    • Analytics
    • UX
    • Redux
    • JavaScript
    • Node.js
    • Material-UI
  5. Software Engineer

    NWSLAB · Full-time

    Nov 2022 - Sep 2023 · 11 mosYerevan, Armenia · Hybrid

    Carpathia UI - Warner Music Group, ADA

    • Worked in cross-cultural environment spanning Russia, Ukraine, Armenia, US, Romania, India
    • Contributed to architectural decision-making with backend engineers and Solutions Architect
    • Transformed Figma designs into high-quality, maintainable code
    • Peer-reviewed code under Tech Lead's guidance
    • Implemented generic TypeScript utility types for project-specific needs
    • JavaScript
    • React
    • TypeScript
    • ESLint
    • Redux
    • Spring
    • Node.js
  6. Software Engineer

    Datafoundry · Full-time

    Mar 2022 - Nov 2022 · 9 mosYerevan, Armenia · Hybrid

    DF mSignal AI - Clinical Trials, Signal Management

    • Kick started a dashboard analytics app, with 4 user roles.
    • Integrated and reused the legacy version of the project.
    • Optimized the performance by modifying the state management.
    • Refactored the codebase following the SOC concept and DRY principle.
    • Designed tech solutions by collaborating with the back-end engineers.
    • JavaScript
    • React
    • Code Review
    • Saga
    • Ant Design
    • Spring
    • Node.js

Education

  • 2025

    Anthropic Academy: Claude with Google Cloud's Vertex AI

    Tool-calling, Agents, RAG, MCP, Vision, Prompt Caching, Claude Code, Computer Use

  • 2018-2019

    Bootcamp - "BPC Armenia" [C++, Python, Linux, Bash]

    Artak Yenokyan - Sr. Engineering Manager at Siemens EDA