Case Study · Software Product · EdTech

BUNGUMAN

An AI-powered flashcard app for learning German — endless, level-appropriate sentences generated on demand, scheduled by spaced repetition so they actually stick. Built solo, from prototype to a shippable, monetised product.

Role

Founder & Developer

Timeline

2026

Domain

Language learning · PWA

Stack

React · Supabase · Gemini

Overview

Where it started

Most flashcard apps hand you a fixed deck — learn it and you're done, and the sentences rarely match how a language is actually used.

BUNGUMAN takes the opposite approach: instead of a finite list, it generates fresh German sentences at the learner's level on demand, and uses spaced repetition to bring each one back at the right moment. The content never runs out, and the review schedule does the remembering for you.

BUNGUMAN app — A German sentence card, generated fresh for the learner's level.

The problem

Flashcard apps run out — and don't manage memory

Before · typical flashcard apps

Fixed decks that run out; nothing new to practise
Sentences that feel canned, not natural
"Review everything" instead of reviewing what you're about to forget
No path to retention — you see a card once and move on

After · BUNGUMAN

Endless, AI-generated sentences at your level
A shared pool that grows richer as people use it
SM-2 spaced repetition surfaces cards right before you'd forget
Dictionary lookups and audio built into the card

How it works

The learning loop

Pick a level

A1, A2, and up.

Get a sentence

Gemini 2.5 Flash generates a fresh German card at that level.

Flip & check

Reveal the meaning, tap a word for a dictionary lookup, hear it read aloud.

Rate it

Easy or Hard.

Scheduled to return

SM-2 sets the next review interval from your rating.

Pool grows

The sentence joins the shared pool for everyone.

Under the hood

The Gemini key never touches the client

The client is a React PWA; everything that costs money or needs a secret lives behind Supabase. Every sentence request goes through an Edge Function, never straight to Gemini.

Client

React PWA

Offline cache

Capacitor (mobile)

→

Backend

Supabase Auth

Postgres · shared pool + progress

→

AI layer

Edge Function

Gemini 2.5 Flash · rate-limited

→

Extras

FreeDictionary API

TTS audio

Stripe billing

The Gemini key never touches the client — every call goes through the Edge Function

Features

What the app does

01 · Generation

AI sentence generation

Gemini 2.5 Flash produces fresh, level-appropriate German — never a fixed list that runs out.

02 · Retention

Spaced repetition (SM-2)

Cards return exactly when you're about to forget them, not on a flat schedule.

03 · Content

Shared sentence pool

Generated sentences land in a shared Supabase pool — a library that grows for everyone.

04 · Comprehension

Dictionary + audio

Instant word lookups via FreeDictionary and spoken pronunciation on each card.

05 · Access

Offline PWA

Installable, with cached cards — study without a connection.

06 · Business

Freemium

A free daily tier, with Pro unlocking unlimited cards, all levels, and full history.

Engineering

From prototype to shippable product

BUNGUMAN started as a single-file prototype with the Gemini key sitting in the front end — fine for a demo, unshippable for real. Getting it release-ready meant rebuilding the foundations: I moved every Gemini call behind a Supabase Edge Function so the key stays server-side, added rate limiting so no single user can drain the quota, and turned on email confirmation to keep fake accounts out.

That's what turned a working toy into something I could charge for.

Monetisation