Skip to content
PortfolioAbidjan

Front-end developer, full-stack leaning

Case study · Personal product

Fassbar, the AI tutor that makes course material understandable

"What was hard to understand becomes fassbar." An AI tutor that starts from students' own course material to create summaries, exercises and explanations, instead of answering off-syllabus.

Visit the live project
Role
Product, design, development, deployment
Team
Solo
Period
2025 – present
Status
Beta with first testers

01

Context and problem

German studies students at Ivorian universities work with dense course material written in German. General-purpose AI tools often answer off-syllabus, or at a language level far from their documents.

The starting question: how do you help a student understand their own course, rather than a topic in general?

02

My role

I carried the project on my own, from research to launch: product definition, design, full development, deployment and first beta testers.

  • 01Product definition
  • 02Design of 41 screens
  • 03Full-stack development
  • 04Deployment and beta

03

Approach

One principle guided every decision: start from the document without being confined to it. The AI stays faithful to the course and clearly flags when it goes beyond it.

Summaries are written in the document's language, with difficult words translated into French. Course themes become clickable to explore each concept.

Fassbar splash screen with the mascot and the tagline "Master your course"
Splash screen
Fassbar sign-in screen: an email field and a "Get a code" button
Passwordless sign-in with a one-time code
Fassbar dashboard with imported courses and shortcuts to corrections and notes
Course dashboard

04

Technical choices

Each choice answers a specific constraint.

  • SvelteKit
  • TypeScript
  • Tailwind CSS
  • Drizzle ORM
  • Cloudflare
  • Claude API
  • Better Auth
  1. 01Clean Architecture

    Domain, application, infrastructure and presentation kept apart, so the product can evolve without a rewrite.

  2. 02Cloudflare D1 and R2

    Database and document storage close to the application, on the same platform.

  3. 03Passwordless sign-in

    A one-time code sent by email: nothing for students to remember, no password to protect.

  4. 04Conversation history

    Sent to the AI with every exchange, so exercises stay consistent from one question to the next.

05

Results

The product went from an idea to a beta used by real students.

sign-ups in 24 hours on the waitlist
110
screens designed before development
41
sprints delivered
5
beta testers
10 to 15

06

What I learned

AI needs context: without the conversation history, exercises ended up contradicting each other. Giving it memory of the exchange fixed the issue.

Choose tools for the runtime from day one: on Cloudflare, I replaced Node.js-oriented libraries with the native Web Crypto API and with unpdf for reading PDFs.

Next project

02E-commerce and dashboard

Yofeii Délices

The online shop of a natural spice brand, and the back office that runs it.

Lead front-end developer · storefront and dashboard

  • Astro
  • React
  • Stripe