Balkan Yildirim

Madison, WI

• Built a reproducible greenhouse-gas emissions data pipeline in Python (Jupyter, DuckDB, pandas) to replace manual, Excel-based sustainability reporting.
• Validated and loaded 995 raw SIMAP emissions records into DuckDB, preserving source metadata, load timestamps, scope values, and data-quality checks.
• Produced dashboard-ready emissions tables by fiscal year, scope, and source category, with reviewed exceptions and exportable CSV/Excel outputs.

Madison, WI

• Built ML pipelines mapping Landsat spectral embeddings to airborne-derived vegetation traits across 290K+ pixel-year observations, bridging spaceborne and airborne remote sensing.
• Benchmarked uncertainty-aware training under leave-one-year-out temporal validation, analyzing how label-uncertainty weighting shifts model targets across 8 biophysical traits.
• Tuned gradient-boosted models (XGBoost) with early stopping and per-trait depth optimization, improving normalized prediction error and mapping spectral-signal limits.

Madison, WI

• Analyzed Cosmic Microwave Background (ACT DR6) and DESI galaxy maps using HEALPix spherical-harmonic methods to probe the large-scale structure of the Universe.
• Constructed galaxy overdensity fields and computed masked/unmasked angular power spectra and CMB–galaxy cross-correlations to isolate cosmological signals from observational effects.
• Developed statistical pipelines in Python (NumPy, Healpy, PyMaster) supporting reconstruction of cosmic velocity fields and structure formation over cosmic time.