Pulse

Overview

Pulse is an internal platform at Northwind Labs that watches every step of the checkout flow for more than 2,000 online stores. It answers one question very quickly: is anyone failing to pay right now?

The challenge

The old alerts were based on error rates, so a silent failure, like a payment button that never appeared, could go unnoticed for an hour. We needed to learn what normal looks like for each store and react when reality drifts away from it.

What I built

• An event pipeline in Go that ingests 40,000 events per second into ClickHouse with under five seconds of lag.
• Per-store baselines that adjust for time of day, campaigns and seasonality.
• An incident view that groups related anomalies, so on-call engineers see one problem instead of fifty alerts.

Results

Median time to detect a checkout incident dropped from 14 minutes to 3. Alert volume fell by 62%, and the on-call team finally stopped muting the channel.