Projects

Here are some of my recent personal projects:

NutriGenie: AI-driven Nutrition Chatbot

NutriGenie is an AI-driven nutrition chatbot developed using the OpenAI API, trained on extensive databases of nutritional information and deficiencies. Users can interact with the bot to receive personalized meal plans, nutritional advice, and information on potential deficiencies. The bot leverages state-of-the-art natural language processing to provide accurate and helpful guidance.

Insightify: Text Summarization Using BART Model

Chrome extension that explores how the BART model, powered by Hugging Face Transformers, revolutionizes text summarization. Leveraging the latest advancements in natural language processing, this project showcases efficient summarization techniques for diverse datasets, offering insights into the future of automated content abstraction.

BrawlMetric: Brawler Strategy App

A comprehensive app for Brawl Stars enthusiasts, featuring real-time recommendations, leaderboards, and fun mechanics to enhance the gaming experience. The app provides tailored brawler suggestions, tracks player achievements, and offers humorous insights into gameplay.

FoodSage: AI-driven Nutrition Recommender System

An advanced culinary solution built using a combination of React, Flask, and artificial intelligence frameworks TensorFlow and PyTorch. The system incorporates user preferences, dietary restrictions, and health goals to recommend meals. It uses the Spoonacular API to provide a real-time data pipeline for AI model feeding, enhancing the accuracy of suggestions. This innovative approach led to a 35% boost in the precision of personalized diet plan generation.

GeoSentiment: Advanced Twitter Sentiment Analysis

An advanced Twitter sentiment analysis tool built with Python. It captures live tweets using Tweepy's Stream API, applies NLP with NLTK and SpaCy, and enriches insights using a machine learning model. The tool also provides geographic sentiment mapping and real-time reports, all while being easily scalable with Docker and Kubernetes.

UIUC Housing Explorer: Housing Web App

This comprehensive housing web application was created using React and Node.js. It offers an immersive platform for users to explore available housing options in the college town area of UIUC. The app features an intuitive user interface and dynamic filtering capabilities, enhancing the search experience.

If you want to see more of my work, take a look at my GitHub!

Research Papers

Extending GrabCuts: Multi-Object Image Segmentation for Enhanced Feature Extraction

This research introduces a novel extension to the GrabCuts algorithm for image segmentation. Traditional GrabCuts techniques, while effective in separating a single foreground from the background, are limited in their ability to handle multiple foreground objects. Our work expands the trimap to a quadmap, allowing for the extraction of more than one feature. This paper details the algorithm's development, implementation, and accuracy analysis, demonstrating significant improvements in handling complex images with multiple foreground objects.

Clarifying Nutrition: Enhancing Transparency and Personalization in Dietary Advice through Retrieval-Augmented Generation

This study introduces NutriGenie, a Nutrition Deficiency-Focused Chatbot AI that addresses the challenge of nutritional deficiencies through personalized dietary advice. Utilizing a state-of-the-art GPT-4 model enhanced by retrieval-augmented generation capabilities and robust data indexing with Llama-Index, NutriGenie draws on reputable sources such as the USDA Nutrient Database and CDC Nutrition Reports for accurate information delivery. By integrating source citations directly into its responses, NutriGenie not only educates users about their dietary needs but also significantly enhances their trust in the advice provided. Evaluation results highlight NutriGenie's superior performance over the baseline model, GPT-3.5, particularly in user trust and accuracy.

Adversarial Machine Learning in Network Detection

This paper examines the use of Adversarial Machine Learning and Generative Adversarial Networks (GANs) to enhance Network Intrusion Detection Systems (NIDS). It utilizes the External Classifier Generative Adversarial Network (ECGAN) model, which aims to improve detection capabilities against novel cyber threats. The research uses the CSE-CIC-IDS2017 dataset for analysis and compares the ECGAN model's performance with traditional NIDS. The findings highlight the potential of AML in cybersecurity, offering directions for future improvements in NIDS.

Measuring Foot Traffic with Bluetooth Low Energy

This research presents a novel application of Bluetooth Low Energy (BLE) for monitoring foot traffic in public spaces. Utilizing BLE broadcasting capabilities and RSSI measurements, the study develops a cost-effective and energy-efficient solution for traffic analysis. The system, comprising a broadcaster and an observer, overcomes challenges like MAC address rotation and broadcasting speed optimization. Extensive testing on different BLE platforms, including the nRF5340 and Adafruit Feather, demonstrates the system's efficacy in various scenarios. This work contributes to BLE technology applications in unconventional settings and has potential in event management, retail, and health crisis scenarios. View on GitHub