🚀 Projects
Bhagavad-GPT: Implemented a customized RAG (Retrieval-Augmented Generation) model centered on the ”Bhagavad Gita,” optimizing output by referencing an authoritative knowledge base. Created ∼3,900 Gita embeddings via an open-source model, stored them in Pinecone, and performed relevancy
searches using cosine similarity, ensuring contextually relevant responses. Developed a Full-Stack application using Flask, Bootstrap, and the AI Agent to provide a seamless user experience
Environment: Langchain, Hugging Face, OpenAI API, Flask, Pinecone, Bootstrap
Assessify:
Developed a multi-agent AI system to create and review exams based on provided material, enhancing the accuracy and relevance of exam questions. Built an application to specify the number of questions, difficulty level, and focus of the output for the AI agent system, resulting in a more tailored and efficient exam creation process
Environment: LangChain, OpenAI, Streamlit, AWS S3
ShareAI:
Engineered an AI-powered photo sharing application, utilizing facial recognition for automatic photo sharing
with all recognized friends. Devised an algorithm that traces through the user’s friend network to automatically share uploaded photos with recognized friends’ accounts on the platform. Received cash prize from judge at Formula Hacks 2024, independently awarded. Environment: React Native, Typescript, Supabase, Python, DeepFace
EyeNavigate:
Enhanced web accessibility with a Chrome Extension that allows users to control scrolling with their gaze using HTML, JavaScript, and WebGazer.js. Optimized scroll amount, timing between scrolls, and scrolling smoothness to ensure readability and prevent dizziness for users.
Environment: HTML, JavaScript, Webgazer.js
PintOS: Strengthened support for PintOS, an OS framework for the 80x86 architecture. Optimized kernel thread scheduling with priority inheritance and aging techniques, enhancing responsiveness. Strengthened user program management with system calls and process control blocks, ensuring efficient resource allocation. Extended file system capabilities, adding support for directories, file operations, and implemented a demand-paging virtual
memory system with eviction policies for efficient memory management.
Environment: Objective C, UNIX, ARM assembly, 80x86
GymShot: Developed a full-stack social media site centered around a community of fitness enthusiasts (a.k.a.
“GymShots”). Implemented secure authentication and real-time cloud database storage with Firebase, ensuring data
privacy and a personalized user experience across Android, iOS, and Web platforms.
Environment: Flutter, Firebase, Dart
Huffman Coding: Created a lossless file compression and decompression tool by converting data into frequency-based codes using a Huffman code tree, decompressing files of any type. Based on Dr. David A. Huffman’s lossless data compression algorithm from “A Method for the Construction of Minimum Redundancy Codes”.
Environment: Java
Machine Learning Projects
Full Stack Lip Reader: Developed a full-stack application showcasing a deep convolutional neural network that can read lips. Model achieved over 97.5% validation accuracy by optimizing CTC loss using advanced data augmentation techniques, transfer learning, learning rate schedulers, and refining model complexity with Convolutional layers, stacked bidirectional LSTMs, Dropout, and Time Distributed Layers. Environment: Python, TensorFlow, Streamlit, Jupyter
Sunspot Predictor: Created Neural Network that predicts number of Sunspots for the upcoming month. The model achieved 14.5 mean average error. Optimized learning rate using LearningRateScheduler. Model included Convolutional layer, stacked LSTMs, Dense layers, and Lambda layers.
Environment: Python, TensorFlow, Keras, Google Colaboratory
Shakespeare’s Sonnet Generator: Created an LSTM network that can generate Shakespeare’s Sonnets. LSTM tokenized a corpus from opensourceshakespeare.org and trained with padded n-grams. Achieved 80% training accuracy with raw model.
Environment: Python, TensorFlow, Keras, Google Colaboratory
Sign Language Classifier: Created Convolutional Neural Network that classifies images of hands into the 26 letters of the English alphabet. Images are from the Sign Language MNIST dataset. Using image augmentation, achieved a validation accuracy of 99.9%.
Environment: Python, TensorFlow, Keras, Google Colaboratory
Facial Recognition: Created a tool that allows users control over the image database used to train a facial recognition model. By customizing the training photos, users can explore how the model responds to specific images and fine-tune its behavior.
Environment: Python, TKinter, OpenCV, Visual Studio Code
MISCELLANEOUS
AP Physics 1 Mechanics Crash Course: Authored AP Physics Textbook: Created and published Advanced Placement Physics textbook on Amazon.
https://www.amazon.com/dp/B08GB25JTC