Projects
LLM Microgrid Agents
Researcher · Apr 2026 – Present
Can LLM agents, one per household, negotiate in plain English to fairly share limited solar and battery power during a grid outage?
Research with Prof. Yongfeng Zhang (Rutgers CS) on whether a population of LLM agents, one per household, can negotiate peer-to-peer in plain English to share limited solar and battery power during a simulated 24-hour outage. In a deterministic 30-household simulation, live agents beat a zero-LLM control by 5.8 points of served load on every clean seed, closing 29% of the gap to a perfect-information oracle, with fairness improving at the same time.
Stock Price Prediction ML
Lead researcher · Jun 2024 – 2025
Does Twitter sentiment actually help LSTM stock prediction? A three-stock, 80K-tweet study says no.
Lead researcher of a peer-reviewed study, accepted by the Journal of Emerging Investigators, testing whether Twitter sentiment improves LSTM stock prediction for AAPL, TSLA, and MSFT. Adding sentiment raised average RMSE by about 32% and contributed under 5% of predictive importance, evidence against the naive integration of social-media sentiment into price models.
NapkinNotes
Co-founder and lead developer · Aug 2025 – May 2026
A full-stack web app that turns raw class notes into organized, searchable study resources.
Co-founder and lead developer of a full-stack web app that turns handwritten scans, PDFs, and Word documents into organized, searchable study resources: OCR with Google Cloud Vision and PyMuPDF, Claude summarization, a social layer with follows and comments, and a student marketplace. 100+ Flask routes, 31 SQLAlchemy models, and OWASP-aligned audit logging. A personal project, not a company.
Phase Spector
Co-developer and designer (3-person team) · Jan – May 2026
Rewind. Strike. Survive. A top-down arcade shooter where your only weapon is retracing your own movement.
Co-developer and designer of a top-down wave-based arcade shooter built in Godot 4.6 around a time-rewind combat mechanic: record about two seconds of movement, then rewind at double speed to damage enemies along your trail and deflect projectiles. Three enemy types, mini-bosses every fifth wave, a chain-kill multiplier, and a persistent top-5 leaderboard. Available to 500+ PDS students.