About Fokusia

*💡 Inspiration * We noticed a critical problem in how people learn today: the preparation takes longer than the learning itself. Students spend hours reading PDFs, extracting key points, and manually creating study materials — time that could be spent actually understanding the content.

We asked ourselves: What if AI could do all the tedious prep work in seconds?

That's how Fokusia was born — a productivity tool that transforms any learning material into an interactive, engaging experience instantly.

** 🛠️ How We Built It ** We built Fokusia as a full-stack application with two main components:

Frontend (React + TypeScript)

TikTok-style vertical scrolling feed for videos and games Responsive design optimized for mobile devices Interactive quiz sheets with real-time AI evaluation Backend (Python Flask)

PDF text extraction using PyMuPDF Integration with Google Gemini 2.0 Flash for: Quiz generation (MCQs + open-ended questions) Educational game content (Fault Finder) YouTube search keyword generation Answer evaluation with star ratings AI content expansion for users without materials ElevenLabs API for text-to-speech podcast generation YouTube Data API for automatic video discovery Smart Token Management

We implemented safety mechanisms to control costs:

PDF sampling: max 10,000 characters regardless of document size Podcast scripts limited to ~100 words (~45 seconds) Estimated cost per upload: $0.003-0.005 🧗 Challenges We Faced

Model Compatibility Issues We encountered a 404 models/gemini-1.5-flash is not found error during podcast generation. Debugging revealed we were using an outdated model name — fixing it to gemini-2.0-flash resolved the issue.

2. Georgian Language Support Ensuring all AI-generated content (quizzes, podcasts, feedback) outputs correctly in Georgian required careful prompt engineering with explicit language instructions.

3. Text-to-Speech Quality ElevenLabs TTS struggled with Georgian text containing many consonant clusters. We solved this by optimizing our prompts to place difficult words at the beginning of sentences, not the end.

4. Mobile Responsiveness Our initial UI had buttons and text that overflowed on small screens. We implemented responsive breakpoints (sm:, md:) to ensure proper scaling across devices.

5. Content Flow Architecture Connecting the user's uploaded content through multiple components (Upload → Store → Feed → PodcastSheet → Backend) required implementing a centralized contentStore for state persistence.

📚 What We Learned Prompt engineering is crucial — small changes in instructions dramatically affect output quality API cost management matters — implementing sampling and limits from day one prevents surprises Mobile-first design should be the default, not an afterthought Georgian NLP has unique challenges — most AI models aren't optimized for it, requiring extra attention Integration complexity — combining multiple APIs (Gemini, ElevenLabs, YouTube) requires robust error handling 🎯 The Result Fokusia reduces what normally takes 2+ hours of study preparation into under 1 minute — embodying the core principle of productivity tools: helping people accomplish more with less effort.