Inspiration
As students at Central University, we noticed a recurring friction: critical information is scattered across static PDFs, outdated notice boards, and disconnected portals like SIP and VCampus. Whether it’s a Level 100 student lost at the Miotso campus or a final-year student stressing over graduation requirements, the "information tax" is real. We built BEVIN Scholar to turn that chaos into a single, conversational interface a "Gemini" that actually knows what’s for lunch at the cafeteria and exactly which prerequisites are needed for the next semester.
UI/UX: Designed with Tailwind CSS and Framer Motion, we focused on a "Mobile-First" glassmorphism aesthetic tailored for students on the move between lectures.
What it does
We implemented a "Static RAG" (Retrieval-Augmented Generation) system, converting the Central University Student Handbook and current timetables into structured markdown to prime the AI’s context.
How we built it
To hit our 7-hour hackathon deadline, our team focused on a high-leverage "speedrun" stack:
Framework: Next.js 15 to handle our unified frontend and backend API routes.
AI Engine: Gemini 1.5 Flash via the Vercel AI SDK, chosen for its massive context window and lightning-fast response times.
Challenges we ran into
Our biggest hurdle was Boundary Control. General LLMs tend to hallucinate or wander into non-academic topics. We had to engineer a strict system prompt to ensure the AI stays within "Miotso boundaries."
Accomplishments that we're proud of
The 7-Hour Sprint: Successfully assembling a functional end-to-end RAG (Retrieval-Augmented Generation) system within a single hackathon session.
Contextual Accuracy: Achieving a high degree of precision in answering campus-specific questions like the exact location of the Miotso library or the specific steps for fee payment which general-purpose LLMs typically fail at.
Predictive Logic: We successfully mapped the Central University course prerequisite tree into a logic-based prompt that helps students visualize their academic journey.
Zero-Latency Streaming: Implementing a smooth, streaming UI that makes the AI feel like a real-time conversation partner, not just a search engine.
What we learned
Data Scoping is King: We learned that the power of an AI assistant isn't just in the model (Gemini 1.5 Flash), but in the quality of the "grounding" data we provide. Formatting the Student Handbook into structured Markdown was a game-changer for accuracy.
The 80/20 Rule of UI: We realized that 80% of student needs can be solved with "Quick Action" buttons. Not everyone wants to type; sometimes they just want to click "Today's Menu."
What's next for Central Mind AI
Live Portal Integration: Our next milestone is connecting directly to the SIP (Student Information Portal) via secure APIs to allow students to check their real-time grades and financial statements within the chat.
Voice Navigation: Implementing a "Voice-to-Campus" mode for students on the move, allowing them to ask for directions or timetable updates hands-free.
Log in or sign up for Devpost to join the conversation.