AI-First Engineering Manager

(AI-First) Engineering Manager

Prague

Groupon connects 42 million customers with local experiences - restaurants, events, wellness and travel - and the million-plus merchants who deliver them. Our mission is to get people offline and into the real world at prices that make it possible. We're an AI-native company in the middle of a platform transformation, moving from a deals marketplace to an experience discovery platform that works for customers and merchants at the same time.

Groupon talks to tens of millions of subscribers through email, push and SMS — and changing a single transactional message takes two weeks, because each one is hard-coded into its own backend service in a ten-year-old stack. Your job is to make it two hours: replace ~120 of those one-off integrations with events a marketer can rearrange themselves in our customer-data platform, and move the whole channel onto Groupon's new Type Script foundation as you go.

You'd lead that turnaround — and still be in the code.

This is a turnaround, not a greenfield. Today the channel runs on a decade-old stack: each message wired by hand into its own delivery service, subscriber data spread across separate systems, a change measured in weeks. You'd move it — piece by piece, without dropping a single send — onto the modern foundation the rest of Groupon engineering is consolidating on:

• From weeks to hours. Replace ~120 hand-built integrations with events a marketer can rearrange in our customer-data platform (Bloomreach) — so changing a message is a config, not a two-week deploy.
• Onto one modern foundation. Groupon's new Type Script platform (Encore, built by ex-Netflix engineers): declare a database or queue in a line of code and it's provisioned and observed for you; changes reach production the same day.
• Why now. Groupon runs on infrastructure built for 2,500 engineers, maintained today by ~260. You'd be on the sharp end of fixing that — for a channel that reaches tens of millions.
• AI is native to the new stack — a shared model gateway and coding-assistant rules wired into every service; agents write real code, and review keeps it safe.

You'll report to Nikash Ray (VP of Software Engineering) and own Managed Channel delivery end to end, leading a small team (~3 engineers) in Bangalore with stakeholders in Prague and London. It's a hands-on role: you run the cadence and stay in the system design and the code. Not for you if you want to review from a distance. Right for you if you like taking a critical, under-invested system and making it modern, observable, and boring to operate.

• The migration is the mission — wire offer and click events into our customer-data platform, retire the hand-built integrations one by one, and move live traffic onto the new foundation without a customer noticing.
• Event-driven reliability — you design the message flow so a double-send is impossible by construction, not by luck.
• AI-first incident response — during a recent traffic spike, an engineer used AI-assisted analysis to surface misconfigurations that only showed under load, shipped the hardening, then turned the diagnosis into a reusable runbook. That's the pattern you'd set.
• Agents doing the toil — a code-quality gate and a PR-watcher keep routine reviews off people's plates, so attention goes to the hard calls.
• The read stakeholders rely on — Prague and London know where the channel stands because you tell them before they ask.

• You've taken over a system nobody could fully explain and left it documented, observable, and boring to operate.
• You've moved live traffic off a legacy system without customers noticing.
• You've designed a messaging flow where double-sending was impossible by construction, not by luck.
• You've led a team four time zones away and it got stronger, not just managed.
• You've made AI agents part of how a team ships — and you can say exactly where they failed.

We'll dig into these live — bring the specifics. (You're still hands-on, so show us production code you've written recently.)

• Day 0–30: the pipeline is mapped end to end and written…