Nihaal Subhash

Hi! I'm Nihaal


This is a brief overview of my education, projects, skills and background
Feel free to reach out to me for opportunities, collaborations, or any discussions related to my areas of interest

Keep scrolling to check out some of my selected projects!

Image Captioning using Transfer Learning

- Designed a deep learning-based web application to generate captions for input images
- Based on an encoder-decoder model and transfer learning
- Used Inception V3 for image feature extraction and an LSTM with GloVe embeddings for caption generation
- Implemented in python using Tensorflow-Keras and NLTK and deployed on the web using Angular for frontend and Flask as backend
- Work published in the International Journal of Information Technology and Electrical Engineering

Photo-App

- Designed a photo-sharing web app inspired by Instagram
- Focused on maximizing accessibility for disabled users
- Enabled users to create accounts, engage with posts, and connect with others accounts
- Implemented using JavaScript, React, Flask, REST, Heroku and PostgreSQL

TED on Demand

- Designed a web interface to create TED Talks-style infotainment videos from user-generated prompts
- Fine-tuned the T5 Text-to-Text LLM for script generation and web scraped content from Wikipedia

Detecting plagiarism using Sentence BERT-based classifiers

- Designed a predictive model to detect automated paraphrasing for plagiarism evasion
- Used sentence embeddings generated by the SentenceBERT and MPNet models
- Achieved a peak accuracy of 89.00%, surpassing the previous state-of-the-art result of 83.36%
- Additionally, designed a smaller, efficient variant of the model that utilized only the first three sentences of the input paragraph, yet achieved an accuracy of 87.64%

Recipe ChatBot

- Designed a contextual AI assistant that parses recipes from allrecipes.com and guides the user through making the recipes in the form of a conversational interface
- Understands navigation utterances, context-based questions and answers user queries
- Implemented in python using RASA, NLTK and Spacy

A statistical analysis of COVID-19 severity and pre-existing medical conditions

- Analyzed the impact of various medical conditions and additional factors on COVID-19 severity and mortality rates
- Developed and implemented diverse classification models, including Neural Networks, KNNs, Weighted KNNs, Naïve Bayes Classifiers, Decision Trees, and Random Forest
- Evaluated and compared the performance of each model in predicting COVID-19 severity and fatality rates, assessing their relative effectiveness through comprehensive performance metrics
- Conducted hypothesis testing using “Student's T-test” to verify the significance of findings

Wildfire Analysis

- Analyzed an extensive dataset of 1.88 million wildfires, incorporating 50 distinct features
- Performed EDA using Tableau for data visualization and OpenRefine for data cleaning and imputation
- Implemented feature engineering using one-hot encoding and min-max scaling through Pandas and Sklearn
- Developed models for predicting the cause of the fire, estimating the size of the fire, and determining whether arson was involved using Histogram-based Gradient Boosting Classification Trees
- Achieved accuracies of 61.8%, 75.1%, and 89.2%, respectively