Hello, I'm
Piyush Hinduja
Hello, I'm
Get to Know More
B.E. Computer Engineering (2023)
M.S. Computer Science
(2025)
Playing Badminton
Trekking
Watching Movies
Hello, hello! Welcome to my portfolio! Hello, hello! Welcome to my portfolio! I'm Piyush Hinduja, an MS CS student at the University of Utah deeply passionate about Artificial Intelligence, Large Language Models (LLMs), and Generative AI. I love architecting efficient AI models using PyTorch, working with Big Data through Python, SQL, and Spark, and building scalable web applications with the MERN stack. Whether it's fine-tuning LLMs, exploring the next wave of GenAI applications, or solving real-world problems with data-driven solutions—I’m all in. If you're looking for someone in these domains, I can assure you—you won’t find a better collaborator! Feel free to explore my skills, projects, and experiences, and don’t hesitate to reach out if you’d like to team up for an exciting project or just chat about the latest in AI over coffee.
Explore My
Browse My Recent
Developed an object detection model to detect and classify various types of road damages, such as cracks and potholes. Implemented and compared three different algorithms: YOLOv5, Faster R-CNN, and SSD. The final solution integrates a user-friendly UI designed with Streamlit, allowing users to upload images and instantly view detection results.
Implemented character-level language models using both LSTM and N-gram approaches to predict sequences of characters based on input sequences. The LSTM model, a recurrent neural network, demonstrated strong predictive performance by accurately predicting the next character in a sequence. Meanwhile, the N-gram model, implemented with Laplace smoothing, effectively estimated the probabilities of character sequences.
Implemented a crucial NLP technique, Dependency Parsing, that analyzes the grammatical structure of a sentence by identifying relationships between words. Designed a Multiclass class classifier that takes the GloVe embedded sentence as the input and outputs the relations between the words of that sentence. Finally, evaluated the models based on their Unlabelled Attachment Score (UAS) and Labelled Attachment Score (LAS) scores.
Designed a pizza franchise's web app which allows the user to order a choice of pizzas from the menu and track their order status online. Utilsed HTML, CSS and Vanilla JS for frontend and NodeJs, ExpressJs and MongoDB for backend.
Created a Deep Learning model to recommend drugs based on patient symptoms using two distinct neural network architectures in PyTorch: an LSTM with GloVe embeddings and a BERT-mini classifier. Attained F1 scores of 77.43% with the BERT-mini model and 70.71% with the LSTM model by optimizing batch sizes and learning rates for effective symptom-based classification.
Know about my
Data Analyst
University of Utah (03/2024 - Present)
Architecting a model to track changes in 10-K filings over time, analyzing forward-looking intensity, document length, and financial tone based on insights from Muslu et al.'s 2015 study. Achieved a 27% increase in forward-looking statement identification by improving data preprocessing and reducing per-filing processing time by 40%. Implementing AI-driven analysis by prompting ChatGPT, comparing AI generated outputs against manually coded algorithms to measure performance and efficiency.
ML Researcher
FLUX Research Group (08/2024 - 04/2025)
Committee Head
ISTE-TSEC (06/2021 - 05/2022)
Web Developer
Exposys Data Labs (06/2021 - 07/2021)
Internet Of Things Trainee
Enovate Skill (06/2020 - 07/2020)
Get In Touch
