Tech enthusiast and passionate programmer with a strong interest in AI and machine learning. Currently pursuing an M.S. in Data Science at Northeastern University, I aim to build impactful AI-driven products that solve real-world problems. With experience in cloud DevOps, scalable systems, and data pipelines, I’m driven to bridge the gap between technology and everyday life. Let’s connect to innovate and make a difference!
2021 – 2023
Developed and deployed AI-powered chatbots to streamline cloud resource management by providing intelligent recommendations and automating resource provisioning across platforms such as AWS and Azure. These solutions enabled users to efficiently select and configure cloud resources, reducing manual effort, enhancing operational efficiency, and delivering a seamless user experience.
Python ServiceNow AWS GraphQL Docker Terraform Jenkins
NEU Course Finder is a Retrieval-Augmented Generation (RAG) chatbot built to simplify course discovery for Northeastern University students. Instead of manually browsing through long course catalogs, users can ask natural language questions and receive accurate, context-aware answers about graduate-level courses. The system combines web scraping, semantic chunking, BGE embeddings, and ChromaDB for fast vector retrieval, with a Hugging Face-hosted LLM generating responses through LlamaIndex. The project demonstrates end-to-end GenAI deployment, from data ingestion to live chatbot interaction.
MangaMatrix is an AI-powered comic generation pipeline that transforms short story prompts into fully illustrated, manga-style comic panels. It integrates large language models for structured story generation and elaboration, followed by image generation using cutting-edge APIs like OpenAI's DALL·E 3. The system automates the entire creative workflow—from natural language input to visual storytelling—by generating detailed panel narratives, translating them into visuals, and compiling the results into downloadable manga-style PDFs. MangaMatrix showcases the fusion of NLP, generative art, and web deployment to democratize comic creation for non-artists and storytellers alike.
This project focuses on fine-tuning a large language model (LLM) to answer health and fitness questions grounded in neuroscience. The training dataset was curated by scraping transcripts from Dr. Andrew Huberman’s YouTube podcasts. Fine-tuning was performed using the Mistral-7B Instruct model with QLoRA, enabling parameter-efficient adaptation.
This project is a multi-modal Visual Question Answering (VQA) system that interprets images and answers questions about them. By leveraging computer vision and natural language processing techniques, the model combines visual and textual understanding to provide meaningful responses.
A novel method for the simultaneous estimation and localization of impact damage in composite sandwich plates using branched neural network. The dataset consists of 25 specimens, with low-velocity impact damage varying in impact energy and impact location.
This study explores the intricate relationship between lifestyle choices, mental health, and academic performance among international students. By analyzing synthetic data based on real-world challenges faced by this demographic, we identify factors influencing both mental well-being and academic success.
This project demonstrates my ability to build a GPT-2 model from scratch using PyTorch. I trained a custom Byte Pair Encoding (BPE) tokenizer on the Wikitext dataset, preprocessed the data, and implemented the GPT-2 architecture, including multi-head attention and feedforward layers. I then trained the model from scratch, monitored its performance using training and validation loss curves, and generated text outputs. This project deepened my understanding of transformer-based models, NLP pipelines, and PyTorch, showcasing my skills in machine learning, deep learning, and natural language processing. The implementation is fully documented and available on GitHub for educational purposes.
