The Problem

Traditional to-do lists can become repetitive, making it difficult to track progress and stay motivated as tasks keep piling up.

The Solution

A 3×3 Bingo-style task manager that turns tasks into a visual game. It supports multiple lists, tracks completed tasks and Bingo combinations, and automatically refills spaces from a task queue.

My Notes

I wanted task completion to feel more rewarding than simply checking boxes. That led me to experiment with a Bingo board, an endless task queue, multiple lists and activity history. The goal was to make everyday productivity a little more fun.

The Problem

Traditional mouse-based puzzle controls can feel less natural, while webcams are often underused as interactive inputs.

The Solution

An interactive 3×3 jigsaw puzzle where players capture a photo using their webcam and use pinch gestures to pick up and move puzzle tiles.

My Notes

Inspired by an Instagram reel, I started wondering if hand gestures could become game controls. I experimented with webcam integration, hand landmark tracking and pinch gestures to make capturing photos and solving puzzles feel more intuitive.

The Problem

Finding recipes usually requires knowing what dish to search for. But what if you only know the ingredients you have?

The Solution

An ingredient-based recipe discovery platform using Epicure-Core food and flavour embeddings to find compatible recipes. When no suitable match exists, it explores generating a new recipe.

My Notes

After reading the Epicure paper on food and flavour embeddings, I became curious about how ingredients relate to one another. Inspired by that work, I created a recipe platform with ingredient-based discovery, adjustable servings, recipe cards, read-aloud instructions, and experimental generation for combinations beyond the existing dataset.

The Problem

Caregivers can find it difficult to distinguish why a baby is crying, whether it's due to hunger, burping, discomfort, tiredness or sleep.

The Solution

An audio-classification system that identifies five types of infant cries using machine learning and deep learning. The proposed CNN achieved approximately 80% accuracy, with Raspberry Pi/ESP32 hardware integration and Docker deployment.

My Notes

Working with a teammate, I experimented with traditional ML and deep learning models, audio feature extraction and spectrogram analysis. We explored CNN-based classification and hardware integration to take the project beyond software. The research was subsequently published as an IEEE paper.

Published as “Infant Cry Classification Using Machine Learning and Deep Learning Techniques.”

The Problem

The challenge was to create a distinctive web game that gave familiar chess mechanics a completely different purpose.

The Solution

A strategy game where players guide a knight to find and escape with a hidden king while avoiding enemy chess pieces. Multiple levels and power-ups add to the challenge.

My Notes

This was my internship project at Jabsz Gaming Studios, where I collaborated with a team of five. We brainstormed different game concepts, eventually turning chess pieces into characters with different roles. I worked primarily on frontend development, building responsive UI components and intuitive gameplay interactions.

The Problem

Understanding how neural networks and Transformer models convert text from one language to another can be difficult without hands-on experimentation.

The Solution

A language translation application using Helsinki-NLP's OPUS-MT models to translate between English, Hindi and Spanish. It includes a Gradio interface and BLEU score analysis to evaluate translation quality.

My Notes

While learning NLP, I wanted to understand what actually happens behind a translation model. I experimented with pretrained Transformer models, text preprocessing and sequence-to-sequence translation, then connected everything through a simple Gradio interface running on Google Colab.