StudyFlow

The StudyFlow Story: From "Big Grammar" to True Understanding The Inspiration

We are a team of three students from the Federal University of Technology, Minna, coming from different departments to tackle a shared struggle. Despite our different fields of study, we all faced the same "PDF Graveyard" problem: folders overflowing with dense textbooks and lecture slides that we promised to read but only "scanned" right before exams.

When you finally open those files, you're often met with academic jargon—what we call "big grammar"—that makes it hard to grasp core concepts quickly. We were inspired to build StudyFlow to act as the ultimate bridge. We wanted to create a tool that feels like having a brilliant senior colleague from any department sitting right next to you, ready to explain complex topics in simple, relatable terms. How We Built It

We built StudyFlow as a high-performance web application designed for deep document intelligence across all academic disciplines.

The Brain: We integrated Google Gemini 1.5 Pro via Google AI Studio. Its massive context window is our "unfair advantage," allowing the AI to process entire textbooks at once so it can answer questions with pinpoint accuracy.

The Stack: The frontend is crafted with React, TypeScript, and Tailwind CSS for a clean, "Linear-style" aesthetic. We used PDF.js for rendering and React Flow to visualize how different topics connect.

The Engine: We implemented a RAG (Retrieval-Augmented Generation) pipeline. Using LangChain, we chunk PDF text and store it in a vector database. This ensures that every answer the AI gives includes a specific page citation from the student's own materials.

The Challenges

The biggest hurdle was contextual precision. We didn’t want a generic chatbot; we needed an AI that knew exactly what was on page 42 of a specific Engineering or Science note. Tuning the RAG pipeline to handle messy PDF structures—like tables and multi-column layouts—required significant trial and error. Additionally, because we come from different departments, we had to build the system to be "domain-agnostic," ensuring it was just as effective at explaining complex engineering formulas as it was at breaking down biological processes. What We Learned

This project taught us the true power of Multimodal AI. We realized that AI shouldn't just summarize text; it should transform the student's relationship with information. By integrating features like auto-generated quizzes and "Explain Like I'm 5" (ELI5) toggles, we learned how to turn a static reading experience into an active learning loop that works for any student, regardless of their department.

Built With

• ai
• css3
• gemini
• html5
• rag
• react
• tailwind
• typescript
• vite