Scrollabus

Doomscroll your syllabus. Let the algorithm teach you.

Comment

The Story Behind Scrollabus

Inspiration

Scrollabus started from a frustration I knew too well: the strange feeling of spending hours glued to a feed, effortlessly absorbing endless short-form content, while school often expected learning to happen through sheer force of repetition and rote memorisation. I kept thinking about that contradiction. If our attention has already been reshaped by modern platforms, why are so many learning experiences still designed as if that never happened?

Part of the inspiration came from my own experience doomscrolling. I was fascinated by how natural it felt to keep consuming content when it was broken into bite-sized, emotionally engaging pieces. At the same time, I’ve always been a little spiteful of the way school can reduce learning into memorising facts for an exam, only to forget them right after. As someone who works in edtech, that tension stayed with me. I wanted to build something that didn’t just make studying more efficient, but made it feel more accessible, more human, and more in tune with the way people already interact with information online.

That idea became Scrollabus: a TikTok-style academic feed where students can upload lecture notes, PDFs, or study material, and watch them transform into short-form educational posts written by different AI study influencer personas. Instead of forcing users into one rigid mode of learning, Scrollabus tries to meet them where they are. Some people want plain-English explanations. Some want worked examples. Some remember things better through jokes, songs, or visual metaphors. So I built six different AI personas, each with its own teaching style, tone, and format, to make the same material approachable from multiple angles.

Lessons

What I learned from building this project is that accessibility in learning is not just about simplifying content. It is also about variety, memory, and emotional design. A concept can feel impossible in one format and suddenly click in another. That shaped a lot of the product decisions. I added interactive quizzes, in-character comment replies, direct messaging with personas, and even a memory-native companion layer that keeps track of long-term learning patterns. The more I worked on it, the more I realised that effective learning is not a one-off event. It is a relationship built over time, with repetition that feels alive rather than mechanical.

How I built Scrollabus

On the technical side, I built Scrollabus as a full-stack system that combines content ingestion, AI generation, social interaction, and persistent learner memory. The main app uses Next.js and Supabase, which gave me a strong base for the product experience, authentication, storage, and database logic. Uploaded materials are parsed from text or PDFs, then passed through a generation pipeline that produces up to 30 posts across the active personas. I used n8n as the orchestration layer to handle workflows for post creation, AI replies, external content ingestion, and scheduling. Different AI tools power different parts of the experience: Gemini for PDF parsing and chat, DeepSeek via Featherless for persona generation, ElevenLabs and Gemini TTS for audio, Kling for video, and Dify plus GMI Cloud for the more agentic teaching and follow-up workflows. On top of that, I integrated HydraDB to store long-term learner memory, so the system can remember what a student struggles with and adapt over time.

One of the most interesting parts of the build was making the product feel less like a static study tool and more like a living platform. I did not want users to just upload notes and receive a summary. I wanted their interactions, like scrolling, saving, commenting, and replying, to actually shape future teaching. That meant thinking about the system not just as a content generator, but as an evolving feedback loop. The feed itself became part of the pedagogy.

Challenges

The biggest challenges came from trying to hold together so many moving parts without losing the core experience. Technically, the project spans multiple AI providers, asynchronous workflows, database-triggered behaviour, and different media formats. Getting those systems to work together reliably was already hard. But the more difficult challenge was product-level: making the experience feel cohesive rather than gimmicky. It is easy to bolt AI onto education in flashy ways. It is much harder to design something that still feels meaningful, intentional, and genuinely helpful to learners.

Another challenge was balancing novelty with trust. A meme persona or a song-based explanation can make learning more engaging, but only if the underlying teaching is still accurate and useful. I had to think carefully about how to preserve educational value while experimenting with playful formats. I also had to confront a bigger question throughout the build: if I am borrowing the mechanics of doomscrolling, how do I do that responsibly? The goal is not to make studying addictive for its own sake. The goal is to use familiar interaction patterns to lower the barrier to learning, especially for students who feel alienated by traditional educational structures.

Conclusion

In the end, Scrollabus is my attempt to reimagine studying as something more adaptive, expressive, and accessible. It comes from a very personal place: my own experience of how easy it is to be pulled into content, and how often school fails to make knowledge feel alive. Building it taught me that the future of edtech is not just smarter models or better summaries. It is designing systems that understand people more deeply: how they pay attention, how they remember, and how they feel when they learn.

Built With

Share this project:

Updates