Building a learning platform with Claude Code

Most agentic coding examples you’ll see are marketing websites. Fair enough, that’s where a lot of Webflow designers start. But the methodology scales far beyond landing pages.

I built One Who Learns, a German learning platform, using the same spec-driven approach I use for client sites. Interactive exercises, a spaced repetition vocabulary engine, AI-generated memory palace images, audio playback, and a Chrome extension for capturing new words while browsing. Not a prototype. A product people use.

This is the case study of how that happened.

The problem

I’m learning German for the B1 citizenship exam. I tried Duolingo, Babbel, flashcard apps, and textbooks. None of them had what I needed.

The apps gamify but don’t teach structure. They don’t explain why “Ich gebe dem Mann das Buch” uses dative for the man and accusative for the book. They don’t prepare you for writing semi-formal emails or giving structured presentations, which is what the B1 exam actually tests.

Flashcard apps are just words on a screen. My visual memory is stronger, so I wanted memory palace imagery: vivid, surreal scenes linked to each word. No app was doing this.

Textbooks have good structure but terrible delivery. Finding audio files means downloading MP3s in 2026. Exercises mean reading, solving on paper, flipping to the answer key. No immediate feedback, no tracking.

I had a clear picture of what the product should be. And I had Claude Code.

What got built

The platform combines the structured progression of textbooks with everything the apps were missing.

Interactive units from A1 to B1. Each unit follows a textbook-style arc: receptive input first (reading and listening), guided grammar practice, then free production. Exercise types include gap fills, sentence ordering, conjugation drills, true/false, multiple choice, dictation, and email writing. All mapped to the Goethe exam format.

A vocabulary engine with memory palace imagery. Every word gets a vivid AI-generated scene. “Der Kopf” (head) doesn’t show a stock photo of a head. It shows a giant patchwork head floating in the sky like a hot air balloon with a castle on top. You see it once and you remember it. Color-coded genders (blue for der, pink for die, green for das) reinforce article associations visually. Verb forms, conjugation tables, and spaced repetition are built in.

Audio that just works. No downloads, no track hunting. Click play.

A Chrome extension. When browsing German sites, one click captures a word, enriches it with AI-generated mnemonics and imagery, and adds it to the spaced repetition queue automatically.

AI writing feedback. For writing exercises, a button copies your text with a prompt to send to Claude, ChatGPT, or Gemini for instant correction.

The workflow

The build followed the same spec-driven development methodology I use for websites. Spec first, then roadmap, then phase-by-phase execution with verification at each step.

The difference from a marketing site: more moving parts. A vocabulary engine needs data structures, state management, and persistence. Interactive exercises need input validation, scoring, and immediate feedback. The Chrome extension is a separate codebase that communicates with the main app.

None of this required me to write code by hand. It required me to think clearly about what each system needed to do, write that into specs and plans, and direct Claude through structured execution.

The parts that took the most iteration: getting the exercise interaction patterns right (how gap fills behave on mobile, keyboard navigation in sentence ordering) and tuning the spaced repetition algorithm. These are product design decisions, not coding problems. The code followed once the spec was clear.

What this proves

If you’re coming from Webflow, your mental model of “what I can build” is shaped by what Webflow allows: marketing sites, CMS-driven blogs, maybe some e-commerce with integrations. The ceiling is the platform.

With agentic coding there is no platform ceiling. The same methodology that builds a landing page builds a SaaS product. The constraint shifts from “what does the tool support” to “can I think clearly about what needs to exist.”

A vocabulary engine with spaced repetition. AI-generated imagery per word. A Chrome extension. Interactive exercises with multiple input types. Audio playback. None of these are “website features.” They’re application features. And they were all built by one person directing an AI agent through a structured process.

The question isn’t whether you’re technical enough. If you can design a user flow in Figma and break a project into sections in Webflow, you already have the thinking patterns. The abstraction just moved up.

Try it

One Who Learns is live. A1 is free: 8 units, all exercises, vocabulary with images, audio. See for yourself what one person and an AI agent can build.