Nerva. Designing and building a zero-to-one internal AI workflow tool for product and design teams.

Nerva canvas linking research, ideation, image, video, and HTML prototype nodes into a visual AI workflow

Confidentiality notice

This work is an internal tool and the source material is limited. To respect that, this case study stays intentionally high-level, focusing on the product idea, the zero-to-one build, and the broader design leadership value rather than implementation detail or internal workflows.

Challenge

As AI tools became more useful across product and design work, one friction point kept showing up: experimenting with workflows still required too much translation between ideas, prompts, tools, and output formats. Nerva was an attempt to make that process more visual, more direct, and more usable for product and design teams, treating AI not as a single chat window but as a structured workflow that mirrors how teams already think.

Strategy

Approach Nerva as a zero-to-one internal tool: something I could define, design, and build quickly to test a stronger interaction model for AI-assisted work. Position it with a single clear line, "Nerva is a visual AI workflow builder for product and design teams," so the audience and intent were unambiguous. Make it real instead of hypothetical, using Claude Code as part of the build process to move from concept to working product faster.

Results

Nerva became concrete enough to present publicly inside Dow Jones as part of Product Community: AI Demos on April 22, 2026, as "Nerva, a visual AI workflow builder for product and design teams." The project moved from side experiment to shareable internal artifact: a working point of view on how AI workflows can be made more legible for non-engineering teams, and a visible example of design leadership through making.

An Image Gen Output panel in Nerva showing a generated photo of a hand holding a phone with a WSJ conversational onboarding chat, beside a contents outline of the concept, the image description, and compositing instructions — **What a node produces.** An Image Gen node turned an onboarding concept into a finished phone screen, with the concept, the image description, and the compositing instructions kept beside it. Nerva treats outputs as things you can inspect and reuse, not text to copy out of a chat.

The story

Nerva is one of those projects I value less for scale and more for what it represents. It was a chance to move beyond abstract AI strategy and build a real tool that embodied a product point of view.

A Nerva canvas titled Content Personalization: three Ideation nodes each fanning out into a row of six Image Gen nodes, every node showing its generated concept image, its selected idea, provider and aspect ratio — **The canvas.** One brief, three ideation passes, eighteen concept renders. The structure of the exploration is visible at a glance, and every branch can be rerun or edited on its own.

The core idea was simple: if product and design teams are going to use AI meaningfully, they need interfaces that help them understand and shape workflows, not just submit prompts and hope for good output. A visual builder creates a different relationship to AI. It makes the structure of the work more visible. It gives people something they can inspect, adjust, and discuss together. And it turns experimentation into a design material rather than a black box.

A Research node opened in Nerva: the user prompt editor with a benchmark brief, and the inspector showing the AI model, a completed status, the system prompt, and an output history with two runs — **Inside a node.** The prompt is editable, the model is a setting, and every run is kept in an output history you can compare and restore.

Two Research nodes connected to an LLM node that is running, three seconds in, with a progress bar and a Stop button — **A run in progress.** Two research nodes feed a synthesis node. You can watch the work move through the graph and stop it at any point.

An HTML Output panel in Nerva showing a generated Financial Data Visualization Gallery: six chart panels for Nvidia revenue, Alphabet R&D spend, Tesla gross margin and others, with preview, code and device-size toggles and a 34,997-character count — **A working prototype.** An HTML node rendered a data visualization gallery from an upstream spec, with preview and code views, device-size toggles, and a Download button.

A Component Specs Output panel in Nerva showing a twenty-thousand-character document titled Dow Jones Data Visualization Component and Pattern System, with a contents sidebar, text size and serif toggles, and Edit, Chat, Share, Copy and Download actions — **A spec you can talk to.** A specs node wrote a full component and pattern system from the guidelines. The output can be edited, shared, or opened as a chat to ask questions of it.

What made the project meaningful as a portfolio piece is that it brought together several parts of my practice at once. It was product thinking, because it framed a clear audience and use case. It was interaction design, because the premise depended on making workflows tangible and understandable. And it was design leadership through making, because instead of waiting for a team or roadmap to validate the opportunity, I built the tool myself and used it to create a shared conversation internally.

Nerva canvas titled WSJ App Personalization: hand-drawn onboarding sketches loaded as Input nodes, each connected to an Image Gen node that has rendered the sketch as a finished WSJ app screen on a phone — **Sketch to screen.** Hand-drawn onboarding sketches go in as inputs; Image Gen nodes return them as app screens in the brand's style. The tool made a whiteboard idea reviewable in minutes.

In that sense, Nerva was not just an internal AI tool. It was a way of testing how design can lead emerging technology work: by making ideas real early, giving teams something concrete to react to, and using prototypes as strategy.

Select highlights

Defined and built Nerva as an in-house tool for internal use.
Positioned it as a visual AI workflow builder for product and design teams.
Used Claude Code as part of the build process.
Presented the work in Product Community: AI Demos as a featured internal demo.
Continued evolving the tool after launch, with evidence of a later published app version.