Featured Projects
Featured Project
Gender Representation and Sentiment in News Media (Instagram Analysis)
Collected 150k+ Instagram posts from major news accounts (e.g., BBC, CNN) and fine-tuned a Vision Transformer to detect gender in images alongside a RoBERTa model for caption sentiment. Analyzed correlations between emotional tone and gender presence to uncover disparities in how men and women are portrayed across political media.
Featured Project
Simulating the Emergence of Homophily Through Selection Dynamics
Designed agent-based simulation to test if selection pressures alone lead to homophily in evolving social networks.
More Projects
Agentic Table Join Selection for AutoML Pipeline
Developed an LLM-based agentic system to identify valuable SQL tables for join operations that enhance target column prediction in Kitana’s AutoML pipeline. The system intelligently selects tables based on historical join data and simulated multi-hop joins, outperforming embedding-based baselines by up to 12.5% in R² gains.
EV2Gym-Driveway: Semi-Centralized EV Charging with Reinforcement Learning
Extended the EV2Gym simulator to create a realistic environment for decentralized EV charging coordination. Trained a PPO agent to minimize electricity costs and grid load under semi-centralized cluster control. Demonstrated cost savings over heuristic baselines with small group sizes and analyzed policy sensitivity to hyperparameters and reward design.
Gender Representation and Sentiment in News Media (Instagram Analysis)
Collected 150k+ Instagram posts from major news accounts (e.g., BBC, CNN) and fine-tuned a Vision Transformer to detect gender in images alongside a RoBERTa model for caption sentiment. Analyzed correlations between emotional tone and gender presence to uncover disparities in how men and women are portrayed across political media.
Simulating the Emergence of Homophily Through Selection Dynamics
Designed agent-based simulation to test if selection pressures alone lead to homophily in evolving social networks.
Restoring Ancient Greek Vases With Computer Vision
Finetuned computer vision models for reconstructing missing decoration on damaged Greek vases with a custom dataset. Compared performance across architectures and restoration tasks.
Tolerance in Polarized Societies: A Game Theoretic Agent-Based Model
Simulated ideological dynamics using game theory to explore how tolerance evolves within polarized societies.
Moisture Mapper: Radar-Based Soil Sensing
Built a non-invasive soil moisture sensing system using millimeter wave radar. Designed the signal processing pipeline and trained machine learning models to predict moisture levels from radar reflections. Developed an interactive visualization tool in Streamlit.
Cyclic Executive Scheduler Framework
Implemented a scheduling framework for cyclic executives in embedded systems, emphasizing predictable task timing.
Custom Automated Forex Trading Bot
Implemented a Dockerized Python trading bot that executes a modified Bollinger Bands strategy. Backtested performance on historical price data to evaluate profitability and risk metrics.
Real Time Operating System
Built a lightweight Real-Time Operating System (RTOS) in C++ with custom memory management, priority-based task scheduling, and inter-task message passing.
Basketball Predictor Deep Neural Network
Trained a deep neural network to forecast basketball outcomes using player statistics and match history.
Spotify Shared Listening Room
Built a full-stack app for synchronized music playback using Spotify API, Django backend, and React frontend.