Independent project Β· deployed on GitHub
CarbonLens β an open-source AI sustainability framework
GPT-3 β 700k liters water Β· cross-referenced against Atlanta datacenter power draw
A low-code tool I built to track the carbon and water tradeoffs of AI datacenters.
It pairs public model-training data with regional utility data on an interactive
dashboard, built with HTML/JavaScript and Power BI.
Research paper
Light Pollution's Mathematical Impact on Astrophysical Observations
Ξm = 2.5 log(32/25 nW/cmΒ²/sr)
Quantified a 0.37-magnitude loss in Atlanta's night sky (a 28% increase) using VIIRS satellite radiance data and a derived limiting-magnitude equation.
Research paper
Power Law Scaling of AI Carbon Emissions with Model Size
COβ β params^1.9
Derived a scaling law from GPT-3 and BERT training data using log-log regression (Ξ± β 1.8β2.0), modeled in Excel.
Science fair experiment
Effects of AI-Level COβ Production on Plant pH & Phosphate Levels
pH 6.8 β 6.1 (spinach) Β· 7.7 β 6.1 (philodendron)
Exposed Malabar spinach and philodendron to 1,200 ppm COβ β datacenter-level concentrations β and measured the resulting drop in soil pH against untreated controls.
National finalist Β· Modeling the Future Challenge
AI Data Center Risk: Carbon Emissions and Water Depletion Threats to Metro Atlanta
Exponential emissions & water-stress index modeling, Python
Built an exponential emissions-projection model and a water-stress index using government-backed sources, comparing mitigated, worsened, and base-case scenarios for Metro Atlanta's freshwater supply.
Submitted Β· Toshiba ExploraVision
Clear-Orbit Skin (COS-Skin): An Orbital Debris Reducer
Passive drag-coefficient coating Β· no propulsion required
Designed a voltage-activated smart coating that raises a satellite's atmospheric drag coefficient to speed up deorbiting, evaluated against drag sails, tethers, and propulsion through a systems-engineering tradeoff analysis.