Strangeworks Aura

Strangeworks formulates hard optimization problems for quantum, quantum-inspired, and HPC solvers. Discovery meant interviewing our own consultants: PhD physicists, a quantum ML scientist, people writing data science code against quantum hardware.

Two findings. The process repeated across clients, so it could be productized. And the make-or-break step was problem definition — that one steered the whole product.

The design question was where to land on a spectrum: a digital science binder at one end, a notebook IDE at the other. Too far toward the binder and scientists can't work. Too far toward the notebook and it's Jupyter with extra steps.

I wireframed across that whole range, then presented the vision at our quarterly offsite in Austin. Having the science team argue with it in person is what made the direction stick.

The dev team jumped into the app building agentic reasoning, and I jumped into the code with them. The app was called Workflows back then — it became Aura later.

Going straight to the front-end is what made the loop fast. Our own ten-scientist consulting team used it while we built, then science teams at Deloitte, Accenture, and Johnson & Johnson built their own models in it — people we hadn't trained, on problems we hadn't scoped. Their feedback landed back in the repo the same week.

Version one, Workflows, was a long linear march. Watching it get used showed us exactly where it broke down.

A year of beta testing showed us where users got stuck, and as the agents got better we jumped ahead of our own roadmap. 2.0 simplifies the whole thing, problem definition through to run.

Our first instinct was an agent per step. 2.0 flips it: the user works back and forth with a problem analysis agent until the definition genuinely holds. That's where the effort goes now. After that, the interface is mostly showing what's happening — formulation, solver selection, data cleaning, compute, run.

This is the end product, building optimization models today. And it's what the consultants told me in discovery two years earlier: the whole thing is that one insight, built.

Aura 2.0 builds the model — then the model has to run a business: staff scheduling, vehicle routing, cargo and freight, logistics.

So the last step is design work again. We're a consultancy, so every engagement ends in its own custom web app, built for that client's operation. Powering them is Strange-UI, a separate design system we built for exactly this: each operation differs, but the pieces repeat — schedules, constraints, scenario builders, solver runs. Strange-UI carries those components and the rules for assembling them, so a new client app starts production-ready instead of from a blank file.

Aura is live today behind the Strangeworks optimization practice — see the shipped product.