EduBox — AI Student Hub

Before I explain anything, I would like to say:

Life gets messy: deadlines change, tasks pile up, and campus life adds variables you didn’t plan for. EduBox is about balance — keeping things manageable so you can do more, worry less. Whether you're a freshman overwhelmed by schedule, or a senior juggling multiple commitments, EduBox helps you stay on top without burning out.

Inspiration

Life and study are full of small frictions: lost notes, missed deadlines, context switches between classes, clubs, and personal life. I built EduBox because I wanted a single, intelligent place that reduces those frictions and restores balance. Rather than another siloed tool, EduBox brings notes, planning, campus info, and AI-driven assistance together so students and busy professionals can stay productive without constant manual juggling. The driving inspiration was simple: make it easy to stay on top of life while preserving time for learning and rest. A multi-agent AI hub that acts as a “locker” for every part of life.

What it does

EduBox is an AI-first student and personal productivity hub that helps you manage learning and daily life by:

Organizing notes and files with semantic search and automatic categorization
Keeping an adaptive planner that suggests and reschedules tasks to avoid overload
Sending timely reminders and helping you recover from backlogs
Aggregating campus resources — events, clubs, dining — into one view
Providing a multi-agent AI assistant to answer questions, plan study sessions, and generate study summaries
Offering analytics to track habits, study time, and progress

In short: EduBox reduces cognitive load, automates routine organization, and helps you make balanced decisions about what to do and when.

How I built it

EduBox was designed as a modular, modern web app with AI-first architecture:

Frontend: Next.js + TypeScript, styled with Tailwind and Shadcn/ui for a responsive, accessible UI.
Authentication: Clerk for secure user sign-in and profiles.
Sync/Storage: Convex for real-time sync and simple serverless data operations.
AI & Agents: Multi-agent orchestration built with CopilotKit patterns, LLMs hosted via Vertex/ Gemini using Tools, and Kendo RAG (retrieval-augmented generation) to ground model answers in user notes and documents.
UI/UX components: MagicUI and Aceternity - powered widgets where to speed up development.
Deployment: Frontend - Vercel setup, Backend - Render, Both having environment-driven configuration for keys and providers.

The architecture emphasizes replaceable connectors—swap the LLM, retrieval store, or calendar provider without needing a rewrite.

Challenges I ran into

Data privacy and context handling: designing RAG flows that give helpful context to agents while preventing overexposure of sensitive data.
LLM behavior and hallucinations: balancing model responses with retrieval and verification so answers are reliable.
Multi-agent coordination: orchestrating several agents (planner, assistant, summarizer) to act coherently rather than conflictingly.
Cost and latency: serving LLM-backed features while managing API costs and keeping UX snappy.
Real-world schedule complexity: mapping human schedules (interruptions, overlapping commitments, deadlines) into automated, trustworthy reschedules.
Authentication and per-user telemetry: maintaining correct access boundaries while keeping a smooth onboarding flow.

Moreover UserContext Flow and Internal Handling of user data served as the biggest challenge. The latency of response and usage of multiple services without a producer/consumer architecture , that is separate microservice architecture posed a problem, that was solved using convex and kendo - thanks to their internal architecture for load handling.

Accomplishments that I am proud of

Built a working RAG pipeline that surfaces user notes to the assistant so responses are personalized and grounded.
Implemented semantic file organization and search that reduces time spent hunting down study material.
Created an adaptive planner that can suggest reschedules and recovery plans when users fall behind.
Established a modular codebase and connector approach enabling rapid experiments with different LLM providers and retrieval stores. Along with tools provided to create events / assignments and future study sessions rightaway from your chat and add to calendar etc.
Delivered a polished, responsive UI with light/dark themes and accessible components. Super duper landing page with very hard try to make the UI responsive.
Prepared solid demo video and gave proper comments in files that make it easier for contributors to get started.

What I learned

Product-first AI needs guardrails: good retrieval, prompts, and fallback logic matter more than raw model size.
User control is essential: automatic changes (like rescheduling) must be visible and reversible to earn trust.
Incremental automation works best: start with suggestions before doing automatic actions for users.
Cost engineering is part of feature design: keep heavy LLM calls for tasks that materially benefit from them.
Testing LLM-driven features requires a mix of unit logic tests and human-in-the-loop validation.
Modular design saves time: keeping clear boundaries between UI, retrieval, and agent orchestration made provider swaps painless.

What's next for EduBox — AI Student Hub

Roadmap priorities to make EduBox more useful and production-ready:

Improved onboarding and initial setup flow so new users quickly import notes from softwares like obsidian/notion/joplin, and see value.
Connectors like Calendar and LMS integrations (Google Calendar, iCal, Canvas/Blackboard) for two-way sync and grade/task ingestion. Zoom and Gmeet Integration, Youtube API Integration and GDrive store Integration.
Offline-first or caching strategies so core features remain usable without connectivity.
Personalization: model-driven study plans and habit recommendations calibrated to user goals and constraints.
Robust monitoring and cost optimization for AI usage (adaptive sampling, caching LLM outputs).
Mobile-first polish (PWA or native clients) and accessibility improvements.
Plugin/connector system so third parties can add data sources or custom agent behaviors. Obviously less challenging and can be done now too but requires architectural changes to create maybe a gateway and interdependencies.
Add testing, CI/CD, and deployment workflows (example GitHub Actions + Vercel pipeline).

Additional Info:

Project Structure:

EduBox/
├── frontend/                 # Next.js application
│   ├── app/                  # App Router pages and API routes
│   ├── components/           # React components
│   │   ├── magicui/         # MagicUI components
│   │   ├── schematic/       # Schematic billing components
│   │   └── ui/              # Shadcn/ui components
│   ├── convex/              # Convex database schema and functions
│   ├── lib/                 # Utility libraries
│   ├── hooks/               # Custom React hooks
│   └── types/               # TypeScript type definitions
├── backend/
│   └── nuclia-sync/         # Nuclia document sync service
└── README.md