What I Built

I built StudyBuddy – AI Companion for My Friend, a responsive, offline-first study dashboard designed specifically for my friend Arun. Arun is preparing for dense technical certifications (specifically the Cisco CCNA & Computer Networks engineering exams) and frequently struggles with three common study bottlenecks:
Passive, high-stress reading of dry, technical documentation (such as RFCs, port mappings, and OSI layers).
The lack of interactive, safe, and personalized practice quizzes that diagnose why an answer is wrong, rather than just grading it.
Managing consistent study patterns and timelines leading up to high-stakes exam dates.
StudyBuddy solves this by converting static lecture slides and notes into an interactive, friendly educational sandbox. It features an AI Tutor that explains complex topics using real-world analogies (e.g., comparing TCP vs. UDP to registered letters vs. postcards), a Quiz Generator, a 3D Flashcard Reader utilizing spaced repetition triggers, an AI Study Scheduler, and a Knowledge Gap Diagnostics Analyzer that maps weak areas and compiles practice questions.

Code

krithikgokuls / studybuddy-local-ai

🦉 Swappable Local AI study companion powered by Ollama & Llama 3.2. 100% private, offline-first exam preparation, quizzes, 3D flashcards, and tutor chat.

StudyBuddy – AI Companion for My Friend

StudyBuddy is a polished, highly responsive, privacy-first AI study companion built specifically to help students and friends prepare for exams and certifications (like the Cisco CCNA) with less stress and absolute confidence.

Designed around open-weight/open-source AI models (such as Llama 3.2), the application maintains a strict Offline-First / Local AI ethos. Study materials never leave your device unless you explicitly opt to route them through custom local servers.

🚀 Key Features

Dashboard: Comprehensive overview of study streaks, daily target goals, mastered modules, and weak diagnostics.
Materials Hub: Upload or paste text/Markdown notes locally with secure local storage containment.
AI Tutor: Real-time conversational tutoring Grounded strictly in your notes, utilizing simple real-world analogies to break down dry documentation. Includes macro triggers: Explain Simpler, Give me an Example, and Test me.
Quiz Generator: Synthesize multi-choice knowledge check questions…

View on GitHub

How I Built It

StudyBuddy is designed to break the vendor lock-in of expensive, closed cloud APIs.

Swappable AI Abstraction & Local Inference The default architecture is optimized to support Llama 3.2 (3B) or Mistral (7B) running locally on the user's laptop through Ollama (http://localhost:11434). We query Ollama's /api/chat and /api/generate endpoints. To handle structured tasks (like constructing quiz structures with incorrect option explanations or creating study plans with duration values), we pass structured JSON schemas and toggle Ollama's format: "json" parameter. Llama 3.2 interprets and parses these rules directly. Robust CORS Handling: Because browsers block cross-origin requests to local services by default, the app detects connection errors and provides explicit setup steps (e.g., launching with OLLAMA_ORIGINS="*" ollama serve) to make local-hosted inference seamless.

UI & Aesthetic Engineering The interface reflects custom, crafted design elements from the Education & Interactive Simulations guidelines: 60-30-10 Color Scheme: Soft, clean off-white canvas (#F8FAFC) representing 60% of background space; warm whites and orange borders (#FAF6F0) making up 30%; and bright terracotta accent buttons (#D96A43) drawing focus to action points. Zero-Pill Metadata: Follows strict typographical discipline, rendering date indicators and categories as clean text with typographic separators (·), avoiding colored capsule chips. Tactile Spaced Repetition: Flashcards incorporate pure CSS perspective transforms (rotateY(180deg)) for a clean, natural 3D flip animation when clicked. Cozy Graphics: Features custom-engineered visual graphics including an academic owl tutor mascot badge (studybuddy_mascot) and a cozy morning study desk header (dashboard_study_header).

Why Does Open Innovation Matter?

Open-source and open-weight AI is revolutionary for education, student privacy, and accessibility:
Absolute Privacy: School essays, draft notes, or code guidelines are deeply personal and proprietary. closed-source cloud APIs require uploading this information to remote servers, exposing private student details. Running Llama 3.2 locally via Ollama ensures Arun's materials never leave his physical hard drive.
True Offline Independence: Arun studies in library basements, on airplanes, and in transit where network connectivity is spotty or non-existent. A closed API fails instantly in these scenarios. Open-weight models running on local silicon turn Arun's personal computer into an offline, high-capacity tutor.
Infinite Customization & Zero Cost: Traditional cloud APIs charge per-token, which rapidly becomes unsustainable for students running thousands of practice quizzes. Open-weight inference is entirely free to run on student hardware, allowing limitless active recall loops.

My Agent Session

I built this entire application in an interactive coding session with the Google AI Studio Agent.

The agent helped me:

Model & Abstraction: Design the swappable AIProvider contract separating the local Ollama module from the offline sandbox.

Tactile UX Design: Scaffold the 3D perspective CSS code for the rotating flashcards and align the colors to a cozy educational theme.

Compilation & Quality: Iterate and compile the React code flawlessly, verifying it with rigorous, zero-error lint checks.

DEV Community

StudyBuddy 🦉 – The Privacy-First, Swappable Local AI Study Companion I Built for My Friend