Building an Open-Source, Multi-Agent Study Companion for My Sister

What I Built

I built Bud AI (Buddy AI)—an open-source, multi-agent educational assistant and interactive study Acompanion designed specifically for college students and my sister.

Studying for college exams often comes with intense mental fatigue, context-switching overhead, and overwhelming syllabus dumps. I built Bud AI to solve three major problems for students:

Context Window Overhead: Breaking down huge study guides into structured, bite-sized academic retrievals using a specialized "Teacher Persona".

Emotional & Focus Tracking: A high-contrast glassmorphic UI that tracks student mood, focus levels, and aura feedback in real-time, providing encouraging visual mascot animations during late-night study sessions.

On-Demand Tool Discovery: Fetching targeted learning resources on-the-fly without leaving the workspace.

Code

jouzia / Bud-AI

Stateful multi-agent AI tutor with memory, adaptive learning, and reward-driven educational benchmarking.

Bud AI x OpenEnv: Multi-Agent Educational Platform & Intelligence Benchmark

A production-ready, multi-agent conversational assistant running on a stateful, reward-driven benchmark environment designed for complex educational task execution.

Application Layer: Bud AI (Next.js / Python Custom Web Interface)
Core Engine: OpenEnv Study Intelligence (Stateful Agent Environment)
Status: Fully Functional & Deployable (Hugging Face Spaces + Vercel compatible)

Bud AI Layer: Glassmorphism Frontend Aesthetic

Bud AI provides a high-contrast, ultra-modern developer interface designed around a strict dark-mode layout:

Brutalist Architecture: Deep near-black backgrounds (#0B0B0F) to minimize cognitive fatigue.
Frosted Interfaces: UI components built as translucent, frosted glass panels with smooth backdrop blur filters and microscopic white borders.
Ambient Signaling: Subtle neon color underglows that visually pulse to signal different multi-agent execution states (e.g., active routing vs. content streaming).

Core Multi-Agent Logic & Routing Architecture

Bud AI transitions past monolithic single-prompt architectures. It utilizes an internal intent routing core that…

View on GitHub

How I Built It

Bud AI is engineered with a modern full-stack architecture built completely around open innovation and flexible AI orchestration:

Frontend: Next.js (React) styled with a high-contrast dark mode glassmorphism interface, custom frosted panels, thin light borders, and responsive UI state management for real-time mood/aura updates.

Backend: FastAPI (Python) serving structured async REST API endpoints.

AI Engine & Agent Framework: Integrated open-weight models and the Gemini API via Pydantic and Instructor for strict schema validation. It leverages a custom Multi-Agent Orchestration pipeline with Model Context Protocol (MCP) concepts to handle tool discovery, context optimization, and persona switching cleanly.

Why Does Open Innovation Matter?

Open innovation and open-weight models give developers true ownership over their agent workflows. For a study tool like Bud AI:

Full Privacy & Customization: Open architectures allow students to inspect how their study data is handled and allow developers to fine-tune models for specific university syllabi.

Zero Lock-In: Unlike proprietary closed APIs that can change pricing or deprecate features overnight, open frameworks let us swap model backends (from open-weight local models to cloud APIs) seamlessly using structured Pydantic schemas.

Community-Driven Extensions: Other students and developers can fork the repo, contribute new agent tools, or adapt the "Teacher Persona" to their own college curriculum.

DEV Community

Building an Open-Source, Multi-Agent Study Companion for My Sister

What I Built

Demo