mateus

Uber

Create and review documents with GenAI

Context and background

Uber is the leader on the ride-sharing business. This case covers the design of the ERD (Entity Relationship Diagram) authoring and reviewing feature, built for internal use by engineers and product managers across the company. The goal was to reduce the friction of creating and reviewing ERDs dramatically, using GenAI (Generative AI) to assist with authoring, surfacing the right reviewers, and cutting down a review process that could stretch over 60 days.

role and scope

user research, user testing, prototyping, documentation, UI design, UX design, vibecoding

stakeholders

1 product manager, 5 engineers, 2 engineering managers, VP supervision

timeline

~5 months from start to finish

tools

Figma, Claude Code

Catch a ride with Uber

What got us here in the first place?

Avg. review time for complex ERDs

60+

days

What the new tool needed

faster review cycles
GenAI-assisted authoring
less manual writing for engineers
live collaboration support

The company is moving toward GenAI solutions and uPlan needed a full redesign to support that direction

“Engineers want to code, not write documentation.”

engineering manager

growth opportunity

Cursor and other GenAI tools are already heavily used at Uber

The problem

The ERD creation and review process at Uber was slow and painful. Complex projects involving legal, privacy, and security faced review cycles stretching over 60 days. The internal tool used for this, uPlan, was outdated and not built for the scale or speed the company needed.

The challenge

Engineers at Uber don't want to write documentation. They want to write code. Any solution had to dramatically reduce the friction of creating and reviewing ERDs, fitting into existing workflows rather than adding to them. The redesign also had to align with the company's broader push toward GenAI-powered tooling, which meant designing for a future state that wasn't fully defined yet.

The opportunity

Uber was already moving toward GenAI solutions across the company, and tools like Cursor were in heavy use. There was a real opening to build a smart, AI-assisted authoring and review experience that could become the standard for how engineers document data systems internally, turning a painful compliance task into something close to effortless.

Desk research and references

How does it work today?

Engineers creating ERDs had to navigate a manual, form-heavy authoring flow inside uPlan. There was no GenAI assistance, no smart reviewer suggestions, and no way to track review status without chasing people across Slack and email. Complex ERDs involving legal, privacy, and security teams regularly sat in limbo for months, with no clear path to resolution or escalation built into the tool.

Who does AI-assisted documentation well?

Cursor is the clearest reference: it doesn't ask you to change how you work, it just makes your current workflow faster and smarter. GitHub Copilot takes a similar approach with inline suggestions that feel like a natural extension of writing code. The common thread is reducing friction without removing control. The engineer stays in charge, the AI fills in the gaps. That's the model this tool needed to follow.

reference images

Research and first iterations

Research

Placeholder — describe what research was or wasn't possible at this stage, and the conditions under which the team had to work.

research type

User interviewsShadowing

time spent

~2 weeks

Objective

Placeholder — explain the reason behind this particular research type, or why a different approach was taken. Sometimes there was no time or space to do formal research.

Results

  • Placeholder finding — what was discovered overall.

  • Placeholder finding — a recurring behavior or pain point.

  • Placeholder finding — a workaround engineers were using.

  • Placeholder finding — something that informed the direction.

First iterations

What was explored

Placeholder — describe the early explorations: sketches, whiteboard sessions, wireframes, or any first-pass ideas that were put to paper or screen.

Result

Placeholder — what came out of this iteration. What worked, what didn't, and what shaped the next step.

Feedback

Placeholder — any feedback received from users or stakeholders at this stage, even informal or directional.

sketches / wireframes

Challenges and steering the ship

Placeholder — name the situation or problem being dealt with in this case.

Challenge 01

The problem

Placeholder — describe the specific challenge, what triggered it, and any feedback that confirmed it was a real blocker.

“Placeholder — a quote or piece of feedback that illustrates the problem.”

The solution

Placeholder — describe what was done to address this challenge. Be honest about whether it fully solved the problem or was a best-effort given the constraints.

solution screenshot or evolution carousel

Challenge 02

The problem

Placeholder — second challenge description and its context.

The solution

Placeholder — how this one was addressed.

solution screenshot or evolution carousel

Results

final screens, collage or video

Final thoughts

Placeholder — overall reflection on the delivery. What the feature became, what it enabled, and any honest assessment of what could have been done differently.

What I've learned

  • Placeholder — a key takeaway from this project.

  • Placeholder — something about process, collaboration, or constraints.

  • Placeholder — something that would be done differently next time.