Atlantis is an AI research automation tool. Unlike Growth OS (which is a workflow tool), Atlantis is a consumption tool. It's designed for analysts who need to read, annotate, and digest massive amounts of information.
The "Perplexity-Style" Interface
The bar for AI interfaces has been raised. Users expect streaming responses—seeing the text appear token-by-token.
I built the Atlantis frontend using Next.js 15 and ShadCN UI. The critical innovation was handling the WebSocket stream from our AI backend.
The Streaming Challenge
When streaming Markdown, standard parsers break if the stream cuts off in the middle of a bold tag (e.g., **bold te).
I implemented a custom buffering strategy that:
- Accumulates tokens in a
Zustandstore. - "Optimistically" renders partial Markdown.
- Smoothly scrolls the viewport as new content arrives (but only if the user hasn't scrolled up to read).
Design System
We didn't want this to look like a dashboard. We wanted it to look like a newspaper. I heavily customized Tailwind CSS typography plugins to create a reading experience that focused on legibility, whitespace, and clear information hierarchy.