×

Financial Research & Sentiment Analysis Tool

A fully local, privacy-first LLM application using FastAPI to securely integrate real-time market data (YFinance) for personal finance tracking and trend analysis.

Fine-tuned advanced Transformer models (Llama 3.1) to parse earnings call transcripts and financial news, extracting linguistic nuances and sentiment scores.

Synthesized quantitative market data with qualitative NLP sentiment analysis to predict market volatility with 82% accuracy, generating secure, actionable investment insights.

×

Algorithmic Trading Bot (Quantitative Finance)

Engineered an automated trading bot integrating the Alpaca API to ingest real-time market data and execute trades based on custom mathematical models.

Constructed and optimized investment portfolios applying advanced mathematical techniques and data-driven quantitative risk management strategies.

Validated model performance through rigorous backtesting across various financial instruments, achieving a 23% simulated portfolio return.

×

NLP – Humor Detection

This Natural Language Processing project focused on detecting humor in movie reviews using a large subset of the IMDb dataset. I implemented transformer-based models for classification, fine-tuning them for humor detection in text. Given the limited GPU resources, I explored zero-shot and few-shot learning techniques to maximize model performance without extensive training. This allowed me to evaluate how transformer models like BART or RoBERTa perform when generalizing to niche tasks. I preprocessed the data for tokenization and cleaning, and implemented evaluation metrics such as F1-score and accuracy. The project challenged me to work with real-world constraints and taught me how to manage large-scale NLP tasks efficiently, optimizing for both resource usage and accuracy.

Link to jupyter notebook: Click Here

×

Purrfect Match (v1)

Purrfect Match was a Spring 2025 team project: a Flask and SQLAlchemy web app that connects pet owners to arrange playdates, meetups, and community events. It had user registration, pet profiles, direct messaging, event RSVPs with email confirmations, and map-based search for compatible pets.

I later rebuilt it from scratch in Django as PetConnect.

Source for v1: GitHub (v1-flask)

×

ZooAdmin

ZooAdmin is an HCI project built in Unity that focuses on zoo administration and animal management. The application combines user-centered design with interactive 3D environments to simulate zoo operations, animal care workflows, and visitor experience. I designed the interface and interactions to explore how spatial UI and direct manipulation can support administrative tasks in a simulated zoo setting.

×

Sandwich

Sandwich is an HCI project developed in Unity that offers an interactive sandwich-building experience. Users assemble ingredients step-by-step in a 3D environment, exploring how gesture-based or pointer-based interaction can make a familiar task engaging and accessible. The project was used to study usability and feedback design in procedural, goal-oriented interfaces.

×

EchoMaze

EchoMaze is an HCI project built in Unity that combines maze navigation with echo-based or audio feedback. Users navigate through a 3D maze where sound and spatial cues guide or challenge the experience. The project explores how non-visual feedback and spatial audio can shape wayfinding and immersion in game-like environments.

×

PetConnect

PetConnect is a social network and playdate matcher for pets, and my Django rebuild of Purrfect Match. Owners share posts and stories, find compatible playmates nearby, and chat in real time.

Highlights: a ranked compatibility score computed in SQL, real-time chat over Ably with per-conversation tokens, a Leaflet / OpenStreetMap map that loads only the visible area, HTMX infinite scroll, a constant query count per page, and a CI-tested Docker deploy on free hosting.

Stack: Django 5.2, HTMX, Alpine.js, Tailwind CSS, PostgreSQL (Neon), Ably, Cloudinary.

Links: Live demo (click "Try the live demo"; the free server may take a few seconds to wake) · GitHub

×

AstroClub

AstroClub is a responsive website I developed for the Astronomy Club using HTML, CSS, and JavaScript. The goal was to improve outreach and communication: members can see upcoming stargazing events, meetings, and announcements in one place. The site includes a homepage, events calendar, and sections for news and club photos, with a mobile-first layout so it works well on phones and desktops. It’s deployed and in use for real-world club communication.

Link to website: Visit AstroClub

×

Pomodoro

I recreated the classic Pomodoro technique in a standalone desktop application using Python and Tkinter. The app provides a simple interface that helps users focus on tasks using timed work intervals followed by short breaks. It includes a timer display, session tracking, start/stop/reset buttons, and a visual countdown bar. Building this project allowed me to strengthen my understanding of Python GUI programming and event-driven development.

×

Pong

I built a version of the classic Pong game using Python's turtle module. The game features two paddles, a bouncing ball, and a scoring system. I designed the interface using turtle graphics and implemented game mechanics like collision detection, velocity adjustments, and paddle control with keyboard bindings. This project helped me grasp core programming concepts such as loops, conditionals, coordinate geometry, and real-time input handling.

×

TFIDF

In this project, I built a Java-based search system for an online library using the TF-IDF (Term Frequency–Inverse Document Frequency) algorithm to retrieve and rank relevant documents. The system included two interfaces—Text and Library—used to compare performance and analyze how different data structures affect search efficiency. I experimented with structures such as TreeMap, HashMap, and ArrayList to evaluate their impact on both runtime and memory usage across various tasks like indexing, term frequency counting, and ranking. The focus was on determining which data structure worked best under each interface in terms of time and space complexity.

×

Agentic RAG (IPL AI Helpdesk)

An intelligent helpdesk system built using an Agentic Retrieval-Augmented Generation (RAG) architecture. It is designed to accurately answer queries regarding Indian Premier League (IPL) data. By leveraging Large Language Models combined with vector databases, the system retrieves precise context and generates highly relevant, context-aware responses. This project highlights my ability to build advanced, domain-specific AI applications that handle real-world information retrieval.

Link to GitHub repository: Click Here

×

Wealth Manager

A comprehensive financial tracking and wealth management application. Built to help users monitor their assets, track expenses, and manage portfolios effectively through an intuitive interface. The project involved designing a structured database architecture and developing a responsive front-end, showcasing my full-stack capabilities in creating functional, data-driven applications.

Link to GitHub repository: Click Here

×

Networks Lab

A collection of practical implementations focusing on core computer networking concepts. This project explores low-level socket programming, network protocol design, and data transmission techniques. Building these labs provided hands-on experience with client-server architectures, concurrency, and understanding how data is reliably transmitted across distributed systems.