Atlas:
automated design tool

As a product designer, I'm supposed to be close to user pain. But the reality is messy. Feedback about design lives everywhere - scattered across Jira tickets, buried in Slack threads, logged in support boards - and none of it finds its way to me in a useful form.

What I usually got was: a teammate pinging me about a complaint someone mentioned in standup. A support ticket forwarded two weeks after it was filed. A Slack message that said "heads up, users are struggling with X" - with zero context on how many, how often, or how serious.

There was no way to see the full picture. No signal that cut through the noise. And by the time a pattern became obvious enough to act on, it had usually already become a fire.

I started wondering: what if all of that feedback - from every source - could be automatically pulled in, classified by an AI, and mapped to the part of the product it actually belongs to?

That's the concept behind Atlas. A design health dashboard where each node represents a product area you own as a designer: onboarding, settings, checkout, search - whatever makes sense for your product. Atlas connects to Slack, Jira, and support boards, processes incoming feedback, and routes it to the right node.

The result is a single view that answers the question designers rarely get to ask directly: *where in the product is friction building up right now?*

The hardest problem wasn't the integrations. It was the information hierarchy.

A dashboard that shows everything equally shows nothing. I needed a visual language that could communicate "this area is healthy - keep going" without it being boring, and "this area is accumulating pressure" without being alarming. The answer was a heat-color system: quiet nodes sit in a calm blue-purple, warming through amber as feedback volume and priority rise, hitting red when something needs immediate attention.

The goal was a dashboard you could scan in three seconds and know exactly where to look - without reading a single line of text.

Classification is where the tool earns its value. A raw Slack message like "users keep getting stuck on the payment screen" isn't useful data - it needs to be mapped to the right area (Checkout), tagged with sentiment (negative), assigned a priority level, and surfaced with enough context to act on.

Atlas uses an AI layer to do this automatically. Every piece of incoming feedback is processed, classified, and routed before it ever reaches the dashboard. Designers don't need to triage - they just open Atlas and the work is already done.

I also designed a manual input path for feedback that doesn't come from an integration: paste a Slack message, a support summary, or field notes from a user interview, and Atlas classifies it the same way.

Atlas started as a personal frustration and turned into a design problem I found genuinely hard and interesting. The most valuable thing I took from it: a good dashboard doesn't present data - it makes a decision for you. Every design choice I made was in service of that: reducing the cognitive load of "where do I look?" to nearly zero.

The project is still a concept. But it shaped how I think about designing tools for knowledge workers - and how much headroom there is for AI to handle the work of connecting dots, so humans can focus on acting on them.