Photo of Dorin Belascu

Dorin Belascu

Senior Product Engineer

9+ years, high ownership across scale and early-stage products

Senior Product Engineer with 9+ years across React, Next.js, and TypeScript. I started in embedded C at Marelli, shipping firmware to millions of production vehicles, which grounded me in reliability, clean architecture, and why maintainability matters.

At Grubhub I take full ownership of features end-to-end on a 1M+ user platform: ambiguous requirements, unreliable data, brittle APIs and all. I use Claude and AI-driven development daily to move fast without cutting corners, and I cofound products on the side with full-stack ownership from idea to production using AI-agent architectures like BMAD.

I own what I build, push back when something won't scale, and ship things that actually work.

Now

What I'm doing right now

Active employment

Grubhub ↗

Senior Software Engineer (Contract)

Since Apr 2024

  • •Auto-trial subscriptions
  • •Personalised offers carousel
  • •Braze SDK major-version migration
  • •Login with Amazon migration

Side projects

nutrichefdelivery.ro ↗

Cofounder & CTO

Since Mar 2025

  • •Meal-delivery platform
efera.ro ↗

Owner, end-to-end IT solution

Since 2025

  • •Storefront on Shopify

Personal projects

Self-directed

Since 2025

  • •reinvent — career-change marketplace
  • •This site

Self-development

Currently focused on:

  • →Learning to become an AI engineer, with Claude as my guide

Work

What I've built, and with what

Every technology below is a filter. Pick one to see the projects it was used on and the problem it solved on each.

Start here
  1. Grubhub ↗

    Senior Software Engineer (Contract)

    Apr 2024 – Present

    Auto-trial subscriptions

    Built the auto-trial subscription feature end to end on a 1M+ user platform — eligibility, enrolment, and the trial-to-paid transition.

    • •20,000+ sign-ups at a ~75% conversion rate.
    • •~$200k in monthly recurring revenue.
    TypeScript
    Kept an eligibility-and-enrolment flow safe to change in a large codebase where the data feeding it was often unreliable.
    React.js
    Composed the trial flow from the components the rest of the app already used, so an entirely new purchase path did not arrive looking like a separate product.
    Redux
    Kept subscription state predictable across a flow spanning several screens and a payment step.
    TestCafe
    End-to-end cover on the purchase path, because a broken subscription flow is invisible to unit tests and expensive in production.
    React Testing Library
    Covered the eligibility branches at component level, where the combinations are cheap to enumerate.

    Braze SDK major-version migration

    Migrated the Braze SDK across a breaking major version on the consumer app.

    • •Zero production incidents.
    Braze
    Upgrade the messaging SDK across a breaking major version without dropping campaign delivery.
    Datadog
    Made the migration safe to ship incrementally — dashboards and monitors proved each rollout stage healthy before the next one opened.
    Progressive Rollout
    Turned an all-or-nothing SDK upgrade into a staged rollout that could be halted at any point.

    Login with Amazon migration

    Owned the cross-platform Login with Amazon migration end to end across web, iOS, and Android.

    • •Shipped on all three platforms with no break in sign-in.
  2. Mar 2025 – Present

    Meal-delivery platform

    Built the product from scratch — storefront, ordering, payments, and the tooling the kitchen runs on.

    • •~1,000 orders and 115 users in the first year.
    • •Scaled to ~30 daily orders through continuous UX iteration.
    Next.js
    One framework for a storefront that had to be fast on mobile and indexable, with the ordering logic server-side in the same codebase.
    Node.js
    Payments, webhooks, and analytics wired up server-side from day one rather than bolted on later.
    Stripe
    Take card payments and handle repeat orders without building and securing a payment stack.
    Sentry
    Catch checkout errors in production before a customer had to report them.
    AI-Assisted Development
    Shipped a production product end to end with a 3-person team, using a BMAD agent architecture.
  3. efera.ro ↗

    Owner, end-to-end IT solution

    2025 – Present

    Storefront on Shopify

    Sole owner of the end-to-end IT solution — a Next.js storefront in front of Shopify.

    • •Running as the live storefront.
    Shopify
    Kept catalogue, checkout, and payments on a platform the business can run without me, while owning the storefront experience in Next.js.
  4. Personal projects

    Self-directed

    2025 – Present

    A full-stack side project connecting career-changers with people already working in the field they want to move into. Built to learn Java, Spring, and the Vue ecosystem at depth rather than reach for familiar tools.

    • •Modular monolith with a REST API, JPA persistence, security, and versioned migrations.
    Java
    Picked deliberately rather than reaching for Node: the backend roles worth targeting are on the JVM, and reading about a language is not the same as shipping in it.
    Spring Boot
    One opinionated backend stack covering the API, security, persistence, and migrations, so the learning went into architecture rather than wiring.
    PostgreSQL
    Relational from the start — the marketplace's core is people, fields, and bookings, which is all joins and nothing document-shaped.
    Flyway
    Versioned the schema from the first commit, so no migration was ever run by hand.
    Docker
    Ran Postgres locally via Docker Compose so the development database matched the production shape.
    Nuxt 4
    Deliberately learned a second frontend ecosystem properly — file-based routing, server composables, and TypeScript throughout — instead of defaulting to React.
    Modular Monolith
    Spring Modulith enforces module boundaries at build time, so services can be extracted later without a big-bang rewrite.

    The site you are reading, with an Assistant that matches job descriptions against this CV and answers questions about my experience — built spec-first, with the domain model and decisions written before any code.

    • •Domain model in CONTEXT.md and decisions in docs/adr/ precede the implementation.
    • •Job descriptions are matched requirement by requirement, citing real Projects and showing gaps honestly.
    Zod
    One schema both constrains the model's Job Match output and validates it before anything renders.
    Vitest
    Integrity tests keep the single data spine honest, and the Assistant is tested end to end with fakes, so the suite never spends API credit.
    Claude Code
    Drove architecture and implementation from a written domain model, so every decision was recorded and reviewable before code existed.
    Claude API
    Answers only from authored CV data; every cited Project is checked against the real data before it renders.
    Upstash Redis
    One store for per-IP rate limiting and an anonymous log of questions the Assistant couldn't answer.
    AI Integration
    Fails safely: hard spend cap, a kill switch, and a fallback that sends visitors back to the page.
  5. Jan 2025 – 2026

    Churn-reduction SaaS

    Co-built a product that connects product feedback to revenue data, so churn signals arrive with a value attached.

    • •Taken from idea to production by a 3-person team.
    Stripe
    Joined qualitative feedback to real subscription revenue, so churn risk could be ranked by the money behind it.
  6. Fortech · urb.io ↗

    React Developer

    Apr 2021 – Mar 2024

    Platform rewrite after a product pivot

    Led a full rewrite of the platform when the product pivoted, and grew the team to deliver it.

    • •Active enterprise clients grew from 3 to 10.
    • •~$50k in additional monthly revenue.
    React.js
    Rebuilt the platform as composable components, so the next product change would not require another rewrite.
    JavaScript
    The platform was plain JavaScript; the rewrite moved it to TypeScript file by file rather than stopping delivery to convert it first.
    Team Leadership
    Doubled the engineering team from 2 to 4 by running the hiring interviews and onboarding, because the rewrite could not land with two people.

    District-heating network features

    Built the map-heavy features end to end for an AI-powered district-heating platform sold to energy companies across Europe.

    • •Network routing, area selection, heat-demand calculation, and ROI screens.
    Mapbox
    Let energy planners draw and route heat networks over real geography, with demand and return on investment computed from what they drew.
  7. Marelli

    Software Project Manager

    Sep 2020 – Apr 2021

    Automotive software project delivery

    Owned an automotive software project end to end, synchronising development, QA, and senior management.

    • •Chosen for the role specifically to bring technical context into project decisions.
    Project Management
    Kept a multi-team automotive delivery on a fixed customer date, tracked through Gantt planning, without losing the engineering detail behind each decision.
  8. Marelli

    Lead Application Developer

    Feb 2020 – Aug 2020

    Core headlamp rendering module

    Led the module responsible for all light output and the innovative lighting features of the headlamp system.

    • •Led a team of 6 developers.
    • •Shipped to production vehicles — the system reached millions of units on the road.
    Embedded C
    Rendered light output in firmware under real-time and safety constraints, where a defect ships inside a car and cannot be patched remotely.
  9. Marelli

    Junior Software Developer

    Aug 2018 – Feb 2020

    Headlamp module features

    Feature work on the headlamp module in an agile team, taking partial ownership of planning and customer discussions.

    • •Mentored a junior team member.
    • •Progressed to lead developer within 18 months.
  10. Hanna Instruments ↗

    Embedded Software Developer (Part-time)

    Apr 2017 – Aug 2018

    Portable measurement meter firmware

    Firmware for portable chemical measurement meters — sensor reading, calibration logic, display, and communication protocols.

    • •Shipped as commercially sold products, including the HI98703 turbidity meter, available worldwide.
    Embedded C
    Turned raw sensor readings into calibrated measurements a lab could trust, on a battery-powered device with only a few kilobytes to work in.

Education

Where I studied

  1. Technical University of Cluj-Napoca

    Computer and Information Sciences, General

    2015 – 2019

Contact

Get in touch

Open to senior engineering roles. Best way to reach me is by email.