ACADEMIC SESSION 2025-2026 COMPUTER SCIENCE PROJECT REPORT Exam Compass Next-Gen Aspirant Intelligence Platform Submitted To: Department of Computer Science Technologies: React 19 * Supabase + Grog Al + Gemini Page1EXAM COMPASS © Project Contributors This project is the result of collaborative engineering, designed and built by the following team members: Name Aditya Shekhar Ayush Kumar Affan Rashid Harsh Bardhan Thakur Shambhav Kumar Jha Narendra Kumar Mission Statement Roll / Contribution Area Full Stack Development & System Architecture Al Integration & Prompt Engineering Frontend UI/UX & Animations Database Schema & Backend Security Testing, QA & Documentation Project Management & Deployment "To democratize high-quality exam preparation in India by bridging the gap between hard work and smart strategy. We aim to replace anxiety with confidence using personalized, data-driven insights accessible to every student."The Challenge Context EXAM COMPASS © User Persona: "Rohan" India sees 1.5 Millions aspirants for JEE/NEET annually. The competition is fierce, and the margin for error is non-existent. 87% Student Burnout Rate Current Gaps v Profi : Class 12 Student. ¥_ Pain Point: Overwhelmed by 50+ chapters across Physics, Chemistry, Maths. ¥_ Need: "I don't know what to study next. keep revising what | already know." Existing solutions (Coaching, YouTube, Books) fail in three critical areas: Static Content Everyone gets the same question bank, regardless of their individual weaknesses. Lagging Feedback Mock test analysis takes hours or days. Students lose the context of their mistakes. Isolation Preparation is lonely. There is no positive reinforcement or gamification to sustain motivationEXAM COMPASS © Our Solution: Exam Compass ‘A comprehensive ecosystem that acts as a Digital Mentor. It doesn't just deliver content; it creates a feedback loop The User Journey FOR THE STUDENT FOR THE EDUCATOR Y Clarity on daily goals. Y Automated question generation. ¥_ Instant doubt resolution via Al. VY Performance tracking at a granular level. Y Gamified rewards (XP/Streaks) for V_ Scalable to thousands of students. consistency. Page 4EXAM COMPASS © Technology Choices & Rationale We selected a "Bleeding Edge" stack to ensure performance, developer experience, and scalability. Frontend Architecture React 19 Selected for its new Concurrent Features CuseTransition’, ‘useOptimistic’). This allows the UI to remain responsive even while fetching complex Al data. Framer Motion Physics-based animation library. Used to create the "Glassmorphic" feel with layout transitions that don't impact browser paint performance. Backend & Al Services Supabase (The Backend) + PostgreSQL: Relational data integrity. + Auth: Secure JWT-based handling. + Realtime: WebSockets for live features. Vite (Build Tool) Replaces Webpack. Uses native ES Modules for near-instant server start and Hot Module Replacement (HMR). Crucial for rapid iteration Tailwind CSS Utility-first approach. Drastically reduces CSS bundle size by purging unused styles in production. Grog Cloud (LPU Inference) Selected over OpenAl for speed. The LPU (Language Processing Unit) architecture delivers 500+ Tokens/sec , making the Al chat feel instantaneous.EXAM COMPASS © System Architecture The application follows a Client-Serverless architecture, minimizing infrastructure costs while maximizing scale. CLIENT LAYER SERVICE LAYER AI LAYER React 19 SPA Supabase Auth oO Google Gemini Zustand Store Postgres DB Liama-3 (Grog) rowser Cache , “ " Edge Functions Data Flow Cycle 1. Interaction: User clicks "Generate Mock Test". 2, State Update: Zustand sets ‘isLoading: true’, UI shows Skeleton loader. 3. Prompt Engineering: Client constructs a structured prompt combining User Preferences (Topic) + System Instructions (JSON Schema). 4, Inference: Request sent to Grog API. Response received in < 800ms. 5. Persistance: Result parsed; typically saved to Supabase (if user saves the test) or held in ephemeral state. 6. Render: React hydrates the component with the new question dataEXAM COMPASS © Feature: Smart Syllabus Tracker Concept IMPLEMENTATION ALGORITHM. A digital twin of the exam curriculum. it tracks | We use a Weighted Moving Average to Not just completion, but "Mastery Level”. calculate the overall subject progress. DATA POINTS TRACKED ¥ Status: Not Started / In Progress / Done / Mastered. ¥ Revision Count: How many times reviewed. Y Confidence: User-rated (1-5 star). Visual Feedback The UI uses color-coded progress bars. Red indicates lagging areas with high weightage (Urgent), while Green indicates mastered low-yield topics (Safe).EXAM COMPASS © Feature: Al Mock Generator The core engine of Exam Compass. It replaces static books with an infinite stream of questions. The Prompting Strategy We don't ask the Al to "Write a question". We give it a Persona and a Constraint Layout. SYSTEM PROMPT LOGIC 1. Role: "You are a ruthlessly strict Professor for JEE Advanced.” 2. Context: "The student identifies as ‘intermediate’ level ¢ "Create a question on ‘Rotational Dynamics’ that tests concept application, not just 4, Format: “Return strict JSON. No conversational filler." Why JSON Mode? Large Language Models (LLMs) can be unpredictable. By forcing JSON mode (available in Llama-3), we ensure that the Frontend always receives a structure it can render, preventing "White Screen of Death" crashes due to malformed API responses.EXAM COMPASS © Feature: Decision Simulator [Link] feature not found in standard apps. It applies Game Theory to exam prepari The Problem The Simulator Students fail exams not because they lack We present Scenario Cards instead of knowledge, but because they panic. They Questions. Users make strategic choices. waste 20 minutes on a 2-mark question. SCENARIO 4: "THE EGO TRAP" "You encounter a Question 15 (Physics). It looks solvable but is taking longer than 4 minutes. It's your favorite topic.” Choice A: Invest 5 more mins | { Choice B: Mark for Review & Skip Al Feedback: "Choice A Is a trap. In competitive exams, ROI (Return on Investment) per minute matters, You chose Ego over Marks."EXAM COMPASS Core Logic: Al Service Layer The ‘sre/lib/ai-ts’ file acts as the bridge between our React frontend and the LLM providers. Note: We chose because it offers the reasoning capabilities of GPT-4 class models but at a latency suitable for real-time interaction.EXAM COMPASS : Global User State Core Logi Managing state in complex apps can lead to "Prop Drilling". We use Zustand for a clean, hook based global store.EXAM COMPASS © Core Logic: Utilities & Helpers ‘Small but crucial functions that keep the codebase DRY (Don't Repeat Yourself) JSON EXTRACTION SAFE-GUARD LLMs sometimes wrap JSON in markdown ('“json ... “"). We need a utility to clean this before parsing. 11 sof Lib/[Link]} ction extractJSON(text: string): any {| sxport Function caloulatexP(questionsSolved: number, aifficulty: string) 4 easy" ‘nedium’ hard’ return questionssolved = (multiplier(aifficutty] Il 10);EXAM COMPASS © Database Schema & Relationships Our Supabase (PostgreSQL) database is designed for referential integrity. 1. PUBLIC PROFILES 2. STUDENT DOCUMENTS Extends the default Auth table. Stores Al-generated revision notes. + id (UUID) - PK + id (UUID) - PK + full_nane (Text) + user_id (UUID) - FK > Profiles + target_exam (Enum: JEE, NEET) + content_json (JSONB) + [Link] (int) + created_at (Timestamp) Security: Row Level Security (RLS) Standard SQL databases are insecure if exposed to the client. Supabase RLS solves this. = Policy: Allow users to only see their own documents REATE POLICY "Enable Read for Users based on User_10"| This ensures that User A cannot fetch User B's mock test history, even if they guess the API endpoint.EXAM COMPASS © UX Design Philosophy Cognitive Load Theory Students are already stressed. The UI must not add to that stress. We employ Cogni Theory to minimize "Extraneous Load” GLASSMORPHISM (THE "WHY") PROGRESSIVE DISCLOSURE The blurred background layers create visual We don't show all 50 features at once. The hierarchy. The content sits "above" the chaos, Dashboard shows only “Daily Tasks". Deeper focusing the eye on the active task. analytics are a click away. Theme Implementation Themes are implemented using CSS Variables and Tailwind. This allows instant Context Switching, ‘text-primary: #226870; —accont-cotor: #004899; /* Neon Pink +]EXAM COMPASS © Algorithm: Smart Document Al How we turn a raw PDF chapter into an interactive Flashcard deck. Pipeline Stages 1. Input Normalization: User uploads text/PDF. We strip whitespace and sanitize HTML characters. 2. Semantic Chunking: The Al breaks text not by pages, but by “Concepts" (e.g., The definition of Torque). 3. Flashcard Synthesis (Gemini 1.5): For each chunk, we ask the Alto generate a pair: { Front: "Question", Back: “Ansner® } 4, Vector Indexing (Future Scope): Storing embeddings of these notes would allow us to “Chat with your Notes" using RAG (Retrieval Augmented Generation). Page 15Security & Optimization FRONTEND OPTIMIZATION ¥ Lazy Loading: Routes (Dashboard, Auth) are split into separate JS bundles. Initial load size < 150KB. ¥ Debouncing: Search bars wait 300ms before triggering API calls to save Token usage. Lighthouse Performance Goals First Contentful Paint (FCP) Cumulative Layout Shift (CLS) Accessibility EXAM COMPASS © ENVIRONMENT SECURITY API Keys (Grog, Gemini) are stored in ‘[Link]’, GitHub Actions CI/CD pipelines use secrets, ensuring no keys are hardcoded in git. 0.8s (Target met via Vite) 0.002 (Visual stability) 98/100 (ARIA labels used)EXAM COMPASS © Future Roadmap We view this version as an MVP (Minimum Viable Product). The vision extends far beyond. @ Phase 2 (Q3 2026) \: Peer Battles Real-time multiplayer quizzes using WebSockets. Challenge friends to a "Physics Duel” © Phase 3 (Q4 2026): Voice Tutor Implementing Whisper (Open for Speech-to-Text. Students can perform *Viva" exams verbally. @ Phase 4 (2027): Mobile App Porting the React codebase to React Native for i0S/Android ecosystem integration.EXAM COMPASS © Conclusion & Impact "Education is the passport to the future." Exam Compass is not just a tool; it is an intervention. By combining the Empathy of a mentor with the Efficiency of Artificial Intelligence, we address the core pillars of successful preparation. Clarity Focus Speed Solved via Smart Syllabus Solved via Gamification Solved via Al Instant Feedback This project demonstrates that high-end educational technology can be accessible, performant, and deeply human-centric. Thank You! Developed with @ by the Exam Compass Team.EXAM COMPASS © References & Citations Libraries & Frameworks v React Documentation: [Link] (Concurrent Mode refs) ¥ Supabase Docs: [Link]/docs (RLS Policies) ¥ Grog Cloud: [Link] (Llama 3 Inference models) ind CSS: v3.4 Documentation Academic Concepts Y Cognitive Load Theory by John Sweller. ¥_ The Forgetting Curve by Hermann Ebbinghaus (Basis for Active Recall features). ASSET CREDITS Icons provided by Lucide React. Fonts served by Google Fonts (inter & JetBrains Mono)