What

Redesign Luca AI’s chat web app so a busy founder can find a past conversation, see what needs attention, and read an answer without scrolling through it.

Context

Luca AI’s web app is a chat interface where founders ask an AI co-founder about their business. Its side menu was a bare history list, with no way to organize or search chats for a user with unlimited questions and limited time. A landing page stood in front of the chat as an unnecessary cover, and status lines like ‘Putting it all together…’ raised more questions than they answered. Responses were long, text-only and double-spaced — one question produced seven sections — and loaded scrolled to the end, sending the reader back up to the top to start.

Enlarge

Enlarge

Enlarge

Enlarge

Enlarge

Enlarge

The original app screens

Constraints

1. Founders short on time. Users could ask Luca anything, as often as they liked, but had little time to sift through the answers — so every screen had to get to the point.
2. A product that was already shipped. New ideas had to be balanced against the app’s existing function and a brand its users already knew.
3. No design system to inherit. Beyond a handful of hex codes there was nothing to build from, so I first set down a basic system — color, type and spacing — and designed each component from scratch on top of it.
4. An 8-hour time difference. With a UK-based team, most of the collaboration happened asynchronously, through recordings, notes and daily updates.
5. No access to users. US clients did not respond to interview requests, and the time difference made sessions with EU clients hard to schedule, so the redesign rests on competitor patterns and reasoning rather than on interviews.

Key Features

1. Folders and chat search

   Chats can be organized into folders and found by search, instead of scrolled for in one long history.

2. Recents within reach

   A recents section keeps the chats a user returns to most one click away.

3. One place for settings

   The profile section at the foot of the menu follows the familiar LLM pattern, and holds both user and business settings.

4. Check-in cards

   The landing opens on insight cards color-coded by priority, each with a ‘Go to chat’ link, so a founder sees what needs attention first.

5. Context before the question

   A row of buttons sets context before a prompt is written, so it does not have to be typed into every one.

6. Progress you can see

   The endless loading loop became a progress bar, keeping the step-by-step checkmarks.

7. Sources on every step

   Each step names the data source Luca is pulling from, giving oversight and a place to start troubleshooting a wrong answer.

8. Answers as cards

   Metric cards show a trend at a glance, as a number and in context, with its source — so an answer needs little or no scrolling.

9. Jump to latest, on request

   An arrow beside the chat jumps to the newest response, instead of the page jumping there on load.

10. Edit, export, speak

    Sent prompts can be edited and resent, responses exported or pinned, and questions spoken rather than typed, for accessibility.

Final Deliverable

Enlarge

Enlarge

Enlarge

Enlarge

Enlarge

Enlarge

Reflection

Studying Claude’s and ChatGPT’s sidebars before drawing my own was the decision that shaped everything after it. Luca’s users are busy founders, not power users of AI tools, and the real risk was a feature-rich menu that turned into a maze. Borrowing patterns they already knew let me save the new ideas for the places Luca actually differs: the check-in cards, the context buttons, and a source on every step.

Handing the redesign to engineers taught me that a design is not finished when the screens are. For each screen type I wrote a short document explaining the decisions and listing what was still open, because an unresolved question in a handoff quietly becomes an engineer’s assumption.

What I would do differently is look beyond Luca’s own client list for people to test with. I would network my way to founders of small eCommerce companies — the market Luca is built for, and the kind of business its current users run — and put a rough prototype in front of a few of them before settling the layout. Choices like how many check-in cards to show, or how short an answer can get before it stops being useful, rest on reasoning and competitor patterns rather than on watching someone use them. The measure I would want most is how long it takes a founder to get from opening Luca to an answer they can act on.