Back to Projects
PYQ Bank cover image

PYQ Bank

Tags
FastAPIReactAIVector Search
Links
GitHub Live Demo
Tech Stack

Backend, Frontend, AI Engine, Data & Search, Styling, Deployment

Overview

Intelligent question bank for NEET/JEE with semantic search and AI explanations.

The Problem

Static PDFs and rigid filters make finding specific practice questions a chore. PYQ Bank introduces semantic search, allowing students to query by concept or difficulty (e.g., 'hard thermodynamics questions'). Powered by vector embeddings and Google Gemini, it provides instant retrieval and deep, step-by-step AI explanations.

What I Did

  • Engineered an Intelligent Question Bank for NEET/JEE aspirants to solve the inefficiency of static PDF practice sets
  • implemented Semantic Search using vector embeddings to allow concept-based querying (e.g., 'hard thermodynamics questions') instead of rigid filters
  • Integrated Google Gemini AI to provide instant, step-by-step 3-stage explanations for every question
  • Built a complete Full-Stack Application with a FastAPI backbone and a reactive React frontend
  • Designed and built a Data Ingestion Pipeline to index questions from Hugging Face datasets into a searchable Whoosh index
  • Deployed the solution with Vercel for the frontend and managed Python backend environments

Tech Stack Details

Backend
FastAPI, Python, Whoosh (Search Engine)
Frontend
React, Vite, TypeScript, Tailwind CSS
AI Engine
Google Gemini API (Generative AI)
Data & Search
Vector Embeddings, Hugging Face Datasets
Styling
Organic Neobrutalism (Tailwind CSS)
Deployment
Vercel

Key Learnings

Full-Stack Integration

Successfully connected a Python FastAPI backend with a React frontend, managing API communication and state

Search Pipeline Architecture

Learned to build a custom search engine using Whoosh, including indexing and schema design for educational content

AI API & Quota Management

Mastered integrating LLMs (Google Gemini) into production apps, handling rate limits (429 errors), and prompting for educational explanations

Dependency Management

Resolved complex Python dependency conflicts for search and AI libraries in a production environment

Design System Application

Applied a consistent Organic Neobrutalism design system across a data-heavy application

End-to-End Product Engineering

Took a concept from 'static PDF' problem statement to a deployed, AI-powered semantic search solution