About Lexicon

Lexicon is an AI-powered vocabulary and exam preparation platform built to make English learning more interactive and personalized for GRE, GMAT, IELTS, TOEFL, and XAT aspirants. The idea came from the need for a single platform that combines vocabulary building, reading comprehension, essay practice, and AI-assisted learning instead of relying on multiple disconnected resources.

How It Was Built

The frontend was developed with React, Vite, and Tailwind CSS, providing a responsive interface with dark mode and customizable themes. The backend uses Node.js, Express, MongoDB, and Mongoose to manage vocabulary data, user progress, and reading passages. Google Gemini AI powers intelligent dictionary lookups, reading passage generation, essay evaluation, and spelling validation, while PDF parsing enables users to import vocabulary directly from study materials.

Challenges Faced

The biggest challenge was making AI responses reliable enough for production. Gemini occasionally returned unstructured text instead of valid JSON, so strict prompt engineering and fallback parsers were implemented to recover malformed responses safely.

Another challenge was implementing the SM-2 spaced repetition algorithm to schedule vocabulary reviews efficiently. The review interval is computed as

[ I_n = \begin{cases} 1, & n=1 \ 6, & n=2 \ I_{n-1}\times EF, & n>2 \end{cases} ]

where the Easiness Factor is updated after each review using

[ EF'=\max\left(1.3,;EF+\left(0.1-(5-q)(0.08+(5-q)\times0.02)\right)\right) ]

Synchronizing review timestamps while preventing duplicate updates required careful backend design.

Maintaining consistent difficulty across AI-generated reading passages was another challenge. A custom complexity score balancing sentence length and CEFR vocabulary levels was used to guide prompt generation.

What I Learned

This project strengthened my understanding of full-stack development, prompt engineering, REST API design, MongoDB schema modeling, asynchronous file processing, and AI integration. I also learned how deterministic algorithms like SM-2 can be combined with generative AI to create a practical and personalized learning experience, while scalable theming with CSS custom properties significantly simplified UI maintenance.