AI Student Assistant

AI Student Assistant puts a verified account system and an AI study companion into a single native desktop window. Students sign up with email verification, sign in, and ask study questions answered by Google's Gemini model through a purpose-built API client. No browser, no hosting, no backend to operate — the entire product ships as one runnable Java application.

Java desktop applications look like Java desktop applications

The brief had two halves that pull in opposite directions. A native desktop app avoids the browser entirely — no hosting, no deployment, no connectivity assumptions beyond the two calls that need it — which is exactly right for a study tool used on a personal laptop. But Swing's stock component set is instantly recognisable and instantly dated, and a student assistant that looks like a 2004 enterprise form does not get opened twice.

Meanwhile the account flow needed genuine email verification, which means driving SMTP from an end-user machine, and the AI layer had to talk to a generative endpoint and survive whatever came back.

• Stock Swing components signal "student project" before a user reads a single word
• Email verification from a desktop client means owning SMTP, delivery failure and OTP lifecycle in-process
• A generative API returns variable and occasionally malformed payloads — the UI cannot assume a happy path
• No browser and no server: everything has to ship inside one runnable application

A custom Swing component layer, a page-swapping frame shell, and an isolated AI client

The interface is rebuilt on top of Swing rather than styled within it. A custom title bar replaces the operating system chrome, and rounded panels, buttons and borders are implemented as reusable components so every screen inherits the same visual language automatically — consistency enforced by the component set rather than by remembering to match the last screen.

A single full-screen frame acts as the shell. Each screen — email entry, OTP verification, signup, login, password recovery and the dashboard — is a separate class producing its own panel, swapped into that frame by the entry point. One frame, one navigation path, no orphaned windows.

Email delivery runs through Jakarta Mail with an HTML template, serving both signup verification and password recovery. The Gemini integration is isolated in a single client class that builds the request payload, posts to the generative endpoint and extracts the response candidate defensively — so a malformed or errored response becomes a readable message in the chat rather than an exception dialog on top of the app.

The problems that took real work to solve. Tap any one to read how.

• A self-contained desktop application covering the full loop: verified signup, sign-in, password recovery and AI-assisted study chat.
• A reusable Swing component layer that gives a native Java application a modern, consistent interface without a third-party UI framework.
• An AI integration isolated behind one client class, so the model or the endpoint can change without touching a single screen.
• A defined next phase: Maven-managed dependencies in place of bundled jars, environment-supplied configuration validated at startup, an embedded database in place of file-based persistence, and unit tests behind a CI pipeline.