All projects
AI022024

NextChat

Real-time AI chat with multi-session support on a high-performance Next.js core.

Lead Engineer Ongoing since launch Self-led at Zytrox Tech 2024

Multi-chat

parallel conversations with persistent history

Streaming

token-by-token responses for perceived speed

Reusable

component system shared across product surfaces

Tech Stack

Next.jsReact.jsNode.jsOpenAI APITailwind CSS

The Challenge

Chat UX lives and dies on responsiveness and state management. Users juggle several conversations at once, expect instant feedback while the model streams, and need history that survives reloads — all on a frontend that must not refetch the world on every keystroke.

Approach

  1. 1

    Designed the conversation model around sessions: each chat is an independent thread with its own message list, so switching chats never loses in-flight responses.

  2. 2

    Streamed OpenAI responses token-by-token into the UI, with an abort path so users can stop a generation mid-flight.

  3. 3

    Built the interface from reusable, themeable components so new chat features (export, editing, regeneration) dropped in without rewrites.

  4. 4

    Tuned Next.js data fetching and rendering so cold loads stayed fast despite a rich client-side experience.

Technical Highlights

Streaming without jank

Buffered token rendering kept the message list smooth under load instead of re-rendering the full thread on every chunk.

State that survives refresh

Session metadata and history are persisted server-side and hydrated lazily, so a reload lands you exactly where you were.

What I Learned

Perceived latency matters more than actual latency in AI UX — streaming plus optimistic UI changes how fast a model feels.
A disciplined component system pays for itself the moment the product roadmap adds a feature.

More Case Studies

Online now

Usually replies within an hour

Chat on WhatsApp1