LearnSynth

LearnSynth converts text/PDF/audio/video into interactive Study Packs (summaries, flashcards, concept maps, quizzes) using the open GPT-OSS 20B model only. Cross-platform: Android, Web, Windows.

Comment

Inspiration

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for LearnSynth## Inspiration

As a second–semester IT student and indie builder, I spend hours jumping between PDFs, lecture notes, videos and voice memos. I wanted one place that could turn any source into a clean study pack, work on Android, Web and Windows, and track progress without friction.

What it does

LearnSynth converts text, PDF, audio or video into a structured study pack using Replicate + GPT-OSS-20B.
It then lets you learn through five modes:

  • Memorization (flashcards)
  • Deep Understanding (guided prompts & explanations)
  • Concept Map (auto-generated relationships)
  • Quiz (quick checks)
  • Cloze Drills (fill-in-the-blank practice)

Your library stores the packs locally and a dashboard shows session time and method-level progress.

How I built it

  • Frontend: Flutter (Dart) targeting Android, Web and Windows with a clean, minimalist UI.
  • Backend: FastAPI (Python) exposing endpoints like /upload-content, /analyze, /study-mode, /review/{id}, /speak, /export.
  • AI Pipeline: Content is transcribed/extracted (for audio/video & PDFs) and sent to Replicate to run GPT-OSS-20B, which returns outlines, concepts, flashcards, quizzes and cloze items.
  • Storage: Local pack storage for offline review, plus a progress model that aggregates sessions and method completion.
  • Deploy: Web on Vercel; API on Fly.io.

Challenges

  • Handling large files and timeouts for long audio/video.
  • Keeping concept maps hydrated when reopening saved packs.
  • Progress sync across study modes so the dashboard always reflects reality.
  • Multi-platform build issues (Android Gradle / Windows desktop packaging / web assets).

Accomplishments

  • A cross-platform app with a full end-to-end AI pipeline from raw content to study pack.
  • Method-level progress and a clean dashboard that students actually understand.
  • Stable library with local packs that you can reopen and continue later.
  • Real demos running on Android, Web and Windows.

What I learned

  • Prompt design and guardrails for open-source LLMs (GPT-OSS-20B) via Replicate.
  • Building once with Flutter and shipping to three platforms.
  • Designing data models for progress tracking and re-hydrating complex UIs (concept maps).
  • Product thinking: simple flows beat feature bloat.

What's next

  • Spaced repetition across all modes and smart review scheduling.
  • Cloud sync & sharing of study packs with classmates/instructors.
  • Export to Anki/CSV, richer analytics and teacher templates.
  • Optional on-device lightweight model for “review on the go” when offline.

Built With

Share this project:

Updates