CODERECET: PassPilot

🔗 Project Repository

👥 Team Name: Stardust Crusaders

Members: Anagha K, Saurav Sreejith, Shreya Padmakumar, S Murugan

---

📘 Project Description

PassPilot is an AI-assisted exam survival guide tailored for KTU students. It optimizes exam prep under tight deadlines using historical PYQ data and topic frequency analysis to generate personalized, minimum-effort study plans aimed at reaching the 40/100 pass threshold.

PassPilot implements scoring algorithms that consider topic frequency, difficulty, and expected yield. The platform supports:

• Survival Track: Minimum viable prep.
• Excellence Track: Extended, deeper study path.
• Crisis Mode: Last-minute planner with an effort-to-marks estimator.

Built on a modular architecture, PassPilot features a real-time pass probability estimator and topic recommendations.

---

🛠️ Technologies & Components

📦 Backend (Engine)

• Language: Python 3.10+
• Frameworks:
  • Flask – REST API
  • LangChain – RAG orchestration
• Libraries:
  • sentence-transformers – Semantic vector embeddings
  • scikit-learn – Cosine similarity
  • numpy – Numerical simulations
  • langchain-google-genai – Gemini LLM + Embeddings
  • chromadb – Persistent vector store
  • pypdf – PDF parsing
  • python-dotenv – Secure env var handling
  • flask-cors – CORS for frontend

🌐 Frontend

• Repo: SauravSreejith/passpilot-web
• Framework: React.js (Vite)
• Language: TypeScript
• Styling:
  • tailwindcss
  • shadcn/ui
• Libraries:
  • react-router-dom – Routing
  • lucide-react – Icons
• Tools:
  • Vite – Build tool
  • npm – Package manager

---

⚙️ Installation

Prerequisites

• Python 3.10+ and pip
• Node.js and npm
• Google Gemini API Key

The project has two repos: backend and frontend. Clone and set up both.

---

🔧 Backend Setup

# Clone the backend repo
git clone <your-backend-repository-url>
cd engine  # or your backend folder

# Create virtual environment
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

# Create environment file
touch .env
# Then edit it with your Gemini key
# e.g., nano .env

Your .env should contain:

GEMINI_API_KEY="AIzaSy...YOUR_API_KEY_HERE"

---

🎨 Frontend Setup

# In a separate folder
git clone https://github.com/SauravSreejith/passpilot-web.git
cd passpilot-web

# Install frontend dependencies
npm install

---

🚀 Running the App

You’ll need two terminals: one for backend, one for frontend.

---

▶️ Start Backend

# In backend directory
source venv/bin/activate
python run_server.py

The backend runs on http://localhost:5000.

---

▶️ Start Frontend

# In frontend directory
npm run dev

The frontend runs on http://localhost:5173.

---

📚 Project Documentation

🎯 Vision: From Anxiety to Strategy

PassPilot tackles study paralysis — the fear of too much to study in too little time. It reframes exam prep as a strategic allocation problem. No more "study everything"; instead: "study smart, survive efficiently."

---

🧠 Component 1: Exam Analysis Engine (ExamAnalyzer)

📌 Semantic Question Analysis

• Model: all-MiniLM-L6-v2
• Goal: Understand question meaning, not just keywords
• Method: Embed both user input and database questions → compute cosine similarity

📌 Pass-Strategy Algorithm

• Formula: Strategic Value = Topic Frequency × Average Marks Per Appearance

• Mechanism:

  • Calculate user’s deficit
  • Rank unstudied topics by Strategic Value
  • Recommend highest ROI topics until gap is closed

📌 Monte Carlo Simulation

• Generate 100,000 simulated papers

• For each:

  • Roll for topic appearance (based on historical probability)
  • Assign random historical mark value
  • Tally up if student would pass

• Output: Pass probability = % of simulations ≥ target score

---

📘 Component 2: RAG Knowledge Engine (RAGAnalyzer)

💡 Pipeline Overview

1. Load & Split: Parse PDFs → Chunk with overlap using RecursiveCharacterTextSplitter

2. Embed & Store: Use embedding-001 → Store in ChromaDB

3. Retrieve: Query embedded → Retrieve top chunks with vector similarity

4. Generate: Use Gemini 1.5 Flash to answer based only on relevant context

✳️ This grounds the LLM’s answers, reducing hallucination.

---

🖼️ Screenshots

• Main Dashboard

• Pass Strategy View

• RAG Chat Interface

---

👩‍💻 Team Contributions

• Anagha K – Frontend Lead Built the entire React frontend with Vite, TypeScript, TailwindCSS, and Shadcn/UI.

• Saurav Sreejith – Backend & Algorithms Built the Flask API, semantic search engine, strategic value algorithm, and Monte Carlo simulator.

• Shreya Padmakumar – Team Lead, UI & Documentation Handled the comprehensive documentation of the project and assisted with frontend design

• S Murugan – AI & RAG Designed the RAGAnalyzer pipeline using LangChain, Gemini, ChromaDB, and PDF processing. Handled DevOps & Architecture