Amazon

Led a 9-person team to consolidate five separate notification-delivery systems into one unified platform.

Shopping notifications reach customers across very different channels — on-device push, classic voice notifications, and ambient cards — and each had grown its own delivery stack over the years. The fragmentation made every new experience expensive to launch and every change risky. I owned the new send path end to end and drove the migration to a single platform that any team can now build on.

The send path was net-new with no in-team precedent, spanning several packages and integrations with externally-owned services whose required guardrails were undocumented and had to be found by diving into each one. Missing them would have silently dropped notifications or bypassed child-device safety protections.

• Designed the delivery path across all three surfaces, with multi-locale and voice support and child-device safety guardrails preserved.
• Integrated directly with the downstream delivery gateway, removing a legacy intermediary service and a network hop from every notification.
• Built the supporting infrastructure and production workflow wiring, and validated end to end with on-device testing.
• Coordinated across the partner teams that owned each dependency, escalating and unblocking where the contracts were unclear.

Brought a flagship shopping feature to Alexa devices by building the backend that powers every product API call.

Add to Delivery lets customers drop items into an upcoming delivery in a single step. It's one of Amazon's fastest-growing shopping experiences — mentioned in the Q3 2025 earnings call and driving billions in annualized sales across the store. My work made Alexa a first-class surface for it.

The hardest part was the product-data layer. I designed the core flow for looking up and modifying items by delivery, co-authored the handler that lets customers change quantities already added, and owned the max-quantity feature — which meant negotiating a retail-side constraint around expensive API calls and documenting the trade-offs so partner teams could align.

• Designed and implemented the product-API integration for all Add-to-Delivery operations.
• Owned the max-quantity design; co-authored the update-quantity handler.
• Delivered the full stack — backend integration through the customer-facing storefront UI (confirmation badges, countdown timers, delivery details).
• Drove API contracts and testing strategy directly with the Checkout backend team across multiple design reviews.

Designed a signal-driven backoff that stops over-notifying disengaged customers — lifting read rate without hurting sales.

A large share of customers had stopped engaging with deal notifications entirely, yet the system kept sending to them — depressing read rates and eroding long-term trust in the channel. I built a mechanism that suppresses Deal Alerts for customers who haven't engaged in 40 weeks, composing cleanly with the existing backoff logic.

Rather than stand up new infrastructure, I inspected the existing pipeline end to end and reused its data store and ingestion path — which inherited its data-deletion compliance and cut the estimated effort from roughly three weeks to one and a half. I owned the full vertical, from the export pipeline to the runtime provider to an 18-case end-to-end test matrix, and launched it as a controlled experiment.

• Designed the backoff signal and the runtime provider that composes with existing suppression logic.
• Reused existing storage and ingestion instead of provisioning new infrastructure.
• Owned the full vertical and an 18-case end-to-end test matrix; launched behind a controlled experiment.

Feature work rides on a healthy fleet — so I carry a real share of the operational load.

I tend to fix the cause beneath a page rather than the page itself: after repeated alerts tied to the same unattributable failure, I added caller-attributed metrics so pages route to the service actually driving load. I hardened a data pipeline against malformed input that had corrupted a stream and blocked roughly a million downstream rows, recovered services crash-looping after framework upgrades, and cleared a large inherited backlog of broken tests and dependency rot.

• Delivered the team's privacy-deletion compliance migration end to end, ahead of a hard regulatory deadline, with a kill-switch and a parallel backstop so no request was lost.
• Took a decayed test suite from a handful of passing tests back to fully green by diagnosing root causes, not skipping.
• Recovered a dead data-sync pipeline and hardened multiple systems against recurring failure modes.