CostGraph.ai

Teams lose track of what they're spending on Kubernetes and VMs. CostGraph surfaces that spend and recommends exactly where to cut.

costgraph.aiSole product designer2026
CostGraph.ai

Intro

Cloud cost tools give you too much data and not enough direction. That is the problem CostGraph exists to fix.

I joined in 2025 as the sole product designer, working directly with engineering to build the platform from the ground up. Everything from UX strategy and the design system to interactive prototypes running on real infrastructure data.

The focus was always the same: show teams what actually needs attention, not everything that could possibly be monitored.

My role

I owned the full product experience, from zero.

  • Product strategy and UX direction
  • User research synthesis
  • Information architecture
  • Design system
  • High-fidelity UI design
  • Interactive prototyping
  • Design-engineering collaboration

All design work was done against real backend exports and infrastructure datasets, not placeholder content.

Process

We started with interviews. DevOps and platform engineers told us what their current tools got wrong, what information they actually looked at, and what they mostly ignored. Those conversations shaped the product direction more than any competitive audit.

From there I used Claude and v0 to move fast through concept exploration. Layouts, interaction flows, visualisation approaches. The goal was to test ideas before committing to them, not to produce polished mockups quickly.

Once concepts felt solid, I built interactive prototypes against real data. That is where the navigation structure, recommendation logic, and workflow decisions got validated. Figma came later, for refinement.

The final phase was about tightening everything. Hierarchy, consistency, interaction states, and preparing flows for implementation.

Design system

Dense technical interfaces go wrong when visual decisions are made ad hoc. I built the CostGraph design system from scratch so that every component, state, and spacing decision had a deliberate answer.

Colour is semantic throughout. Warnings, inefficiencies, healthy resources, each has a distinct visual treatment. Not decoration, just signal.

The system covers the full token set, component library, spacing rules, table patterns, recommendation cards, and interaction states across the platform.

GraphAI

Most AI features in product tools live on a separate page. You go there, ask something, come back. The context is lost.

GraphAI is designed differently. It stays accessible throughout the platform, and the prompts shift based on what the user is currently looking at. Infrastructure insights, anomalies, recommendations. All reachable without breaking the workflow.

It does not announce itself. It is just there when you need it.

Lessons

Clarity beats volume. Infrastructure teams already have access to every metric. What they do not have is a clear answer to "what do I fix first." That is the actual design problem.

Working against real backend schemas changed how I made decisions. When the data is real, you stop designing for hypothetical states and start designing for what actually happens. The product got better because of it.

Next project Luotain