RUBAR OZAN BARSAL

I build the systems that let teams ship software with AI.

Staff software engineer, more than a decade building end-to-end platforms. I introduced the AI-assisted workflow a team develops on — and build on the platform alongside them: integrations, internal tooling, architecture.

Get in touch See the work

AI-ASSISTED DEVELOPMENT
FORWARD-DEPLOYED DELIVERY
UTC+3
REMOTE OR EU RELOCATION

How I work with AI

The interesting part isn’t writing code faster. It’s the workflow a whole team develops on.

Claude Code wired into Jira and Figma over MCP, GitHub CLI automation, custom skills for auditing, documentation and component generation, and layered repository rules with maintained memory — rules that make the coding agents generate against the design system’s components and tokens. Work moves from ticket to pull request through plain language and a few links.

It holds up on Adobe Experience Manager, a stack AI models have seen very little of. It wasn’t the first attempt: Cursor and GitHub Copilot’s agent mode were tried on the same codebase and dropped — neither was performant enough, and both needed more human input than Claude Code to reach production quality.

I introduced it to the main repository in March 2026 as two rule files, and I review its changes as the team extends it: 58 files now, edited by 13 engineers, seven of them in the month to September 2026.

When it works, the constraint moves. Pull request volume roughly doubled and median time to merge went from 3.3 hours to 2.7. The number of changes merged without a human review doubled with the volume — review capacity, not authoring speed, is the constraint the tooling now has to address.

Selected work

Enterprise platform
Adobe Experience Manager

An enterprise AEM platform, and the AI-assisted workflow the team develops on. Client confidential.

Joined a build already in flight; little had shipped
Components built for the platform’s token-based, two-theme design system
Okta authentication integration owned; Stripe payments and a multi-step transactional flow alongside the team
10–15 active developers at a time
Cursor and GitHub Copilot’s agent mode tried and dropped before this one held

I brought agent-based development into the main repository in March 2026 as two rule files, then reviewed the changes as the team built the set out to 58 — 13 engineers editing it, seven of them in the month to September 2026 — and built on the platform through the same workflow, mostly its React front end, Java where needed. Within six months, 25 of the 51 engineers who had committed to the repository were shipping AI-assisted work.

Repository contributors shipping AI-assisted work within six months: 25 of 51
Time for the team to double: 26 weeks
Commits per active developer-week over the same period: +24%
Pull request volume doubled; median time to merge: 3.3 h → 2.7 h

Adobe Experience Manager
Claude Code
MCP
GitHub CLI
Java
TypeScript

Forrester Research
Authoring pipeline

An authoring pipeline that ran on paste.

Seventeen months on the client-facing application that delivers Forrester’s research to its clients (Vue/Nuxt)
The research page owned end to end — a preview for members without access, the full paper with its charts and graphics for those with it, driven by Contentful
The application’s design system built and maintained in code: an atomic component library with SCSS tokens and utilities
Authors used to copy approved documents into the CMS by hand and rebuild the formatting

I built much of the companion Contentful authoring editor — a Word research template goes in, structured rich text and components come out.

Contentful
Vue
Nuxt
React
TypeScript
Design systems

Also worked with

Clients I’ve built for, across finance, insurance, research, hospitality, automotive and government — some as sole developer on campaign work, some as the engineer or tech lead on platforms whose architecture and delivery I owned.

About

More than a decade in agencies, delivering for finance, insurance, automotive and research clients — from small sites to enterprise platforms, as developer and tech lead. The work I’m best at is end to end: architecture, integrations, and the standards a team builds against.

Since 2025 the centre of it has been how a team develops with AI: introducing the workflow, reviewing what the team makes of it as they extend it, and building on the platform through it.

I’ve spent my career in rooms with clients, authoring teams and stakeholders — weekly demos for years — translating between what people need and what actually gets built. Building the tooling a team develops on is the work I want more of.

I studied American Culture and Literature before I wrote code professionally, which is probably why I end up doing as much explaining as engineering. Early on, a faux-3D YouTube masthead I built for a bank’s sponsorship of an Anish Kapoor exhibition got me invited to Google’s Istanbul office to explain the technique to a visiting US team. I work best where communication is direct and people are treated with respect.

Skills

AI-assisted development: Claude Code
Cursor
GitHub Copilot
MCP servers and integrations (Jira, Figma, Slack)
Agentic workflows and guardrails
Context engineering — repository rules, custom skills and memory
Human-in-the-loop review
Rolling tooling out to a team and measuring adoption
Forward-deployed client delivery
Developer experience & tooling: Internal developer tooling
Onboarding
Code review at scale
CI/CD (GitHub Actions)
Infrastructure as Code (Terraform)
IAM
AWS (Lambda)
Google Cloud Platform
Cloudflare
Languages & frameworks: TypeScript
JavaScript
React
Next.js
Vue
Nuxt
Node.js
Express
Java
Enterprise & CMS: Adobe Experience Manager (Sites; certified)
Contentful
Strapi
Front-end craft: Design systems
Design tokens
Component libraries
Accessibility
Performance
Integrations: Okta
Stripe
REST
GraphQL
Languages: Turkish (native)
English (native-level)

Contact

info@obarsal.dev