An experiment in development-first design by creating a LLM chat client

While chat interfaces have been a fascinating and useful shift recently, I felt we had lost an experience as well. The experience of more idly browsing the web, especially a site like Wikipedia, felt casual and exploratory without feeling like a live discussion.

I wanted to create a chat client that re-enabled this workflow without losing the strength of a LLM client.

Development

To create NextQ I skipped the normal design process altogether. I began with a written brief with the core feature set, essentially doing a lightweight design doc. I then immediately fed that into Cursor to get to a functioning prototype as quickly as possible.

My philosophy for prototyping and design is to do whatever is needed to answer the key questions. This could be research, visual design, or less functioning prototypes. But in this case the question was one of exploration with real data, so a fully functioning prototype was the only answer.

By using Cursor I iterated on the full interaction design (outside of a small button design in Figma) to reach a state that felt testable and releasable.

Other details

• Under the surface I also developed a feature for NextQ to summarize the ever changing conversational subject matter, in a high level topic and also sub-topics. This is done to feed context into the question prompting in a lightweight manner, avoiding having to shove lengthy conversational responses back to the API.

• This was my first project using Vercel to launch a small project completely independently. It wasn't as scary as I thought it'd be! Especially with automatic deployment from Github pushes.

YouTube video

Here's my YouTube that provides a more visual overview on the project.