StudyMate

Progress Section(Showing the quizes taken and scores)
The AI Studio
Daily Recap(Shows up between 8pm and 11:59 pm only)
AI Quiz Master(Upload/Select file then choose topic, difficulty level and number of questions)
AI Tutor(Ask any thing from the choosen file, upload Images, give role to AI)

Inspiration

Education is broken. Students today have access to infinite information (PDFs, textbooks, slides), but they lack two critical things: Engagement and Consistency.

Most "AI Tutors" are just wrappers around ChatGPT—they give you the answer, which actually prevents critical thinking. Furthermore, students struggle to organize their study schedules and often repeat the same mistakes because they lack personalized review mechanisms.

I didn't want a simple "Homework Helper." I wanted a Complete Learning Ecosystem.

That’s why I built The AI Tutor: a platform that combines active learning (Socratic Chat), rigorous testing (Feynman Mode), and habit-building tools (Heatmaps & Podcasts) into one seamless experience.

What it does

The AI Tutor transforms static PDF textbooks into an interactive, full-cycle learning platform. It consists of three main pillars:

The Intelligent Chat (Active Learning) Gone are the days of passive reading. Our RAG-powered chat offers distinct "Learning Modes":

Socratic Mode: The AI asks guiding questions to help the student discover answers themselves.

Feynman Validator: The student explains a concept, and the AI grades their understanding (0-100), spotting misconceptions.

The Student Dashboard (Consistency & Planning) Success is about showing up. I built productivity tools directly into the app:

Smart Calendar: A dedicated space to mark Exams, Quizzes, and Revision dates, keeping the student's timeline clear.

Consistency Heatmap: Inspired by GitHub, this visualizes daily study activity. It gamifies the process students don't want to break their "streak," motivating them to study every day.

The AI Studio (Retention & Review) This is our innovation lab for content generation:

The Mini Podcast: A personalized audio review feature. The AI analyzes the wrong answers from yesterday's quiz and generates a short, conversational audio explanation for the student to listen to the next morning. It turns commute time into revision time.

Flashcards (Coming Soon): Automated generation of spaced-repetition cards based on chat history.

How I built it

I built a highperformance architecture focused on accuracy, cost-optimization, and user experience:

Backend: Python & FastAPI for high-concurrency request handling. AI Orchestration: LangChain for managing complex prompt chains and memory. Authentication: Clerk for secure, seamless user management and session handling. Database: Supabase (PostgreSQL + pgvector) for storing chat history, vector embeddings, and user progress data. Vision: Azure Computer Vision to analyze diagrams and charts in uploaded PDFs. Frontend: React.js with TailwindCSS for a responsive, modern interface.

The RAG Pipeline I implemented a sophisticated Context Aware pipeline:

Intent Classification: A heuristic layer detects conversational inputs to skip expensive API calls.

Query Rephrasing: I rewrite follow-up questions (e.g., Why?) into standalone search queries to fetch the right data.

Strict Context Guarding: The AI is strictly bound to the PDF context to prevent hallucinations.

Challenges I ran into

The Over-Helpful Rephraser: Initially, our Rephraser model tried to answer questions instead of just searching for them. I fixed this by implementing strict System Prompts that forced the model to act solely as a keyword extractor.

Connecting the Podcast Logic: Generating the Mini Podcast required complex logic fetching yesterday's specific mistakes, retrieving the correct context again, and synthesizing a friendly audio script without hallucinating new facts.

Token Optimization: To keep costs low, we implemented a Greeting Guard (skipping RAG for small talk) and limited context retrieval to the top-3 most relevant chunks. Accomplishments that we're proud of The Mini Podcast Feature: Creating a system that proactively teaches you about your past mistakes via audio is a game-changer for retention. Multi Modal Capabilities: The ability to understand text, images (diagrams), and generate audio makes the app accessible to all types of learners. AI Performance Coach: We integrated an always-available audio mentor that goes beyond academics, offering real-time guidance and psychological support to keep student morale high and combat burnout. The Feynman Engine: Successfully building a logic flow that acts as a Strict Grader rather than just a "Helpful Assistant."

What's next for The AI Tutor

Expanded Studio: Launching the automated Flashcard Forge, Summary Notes, and Interactive Mind Map generator within the Studio to transform raw text into visual study assets instantly. The Exam Predictor Model: Allowing students to upload past year question papers alongside their textbook, enabling the AI to statistically analyze patterns and predict which topics are most likely to appear in the upcoming exam.