Projects
Question Answering System with BERT, LLMs & Passage Retrieval
Benchmarked four Hugging Face transformer models on factoid and confirmation QA, with entity-level error analysis, LLM-based answer refinement, and BM25 vs. dense passage retrieval.
Photovoltaic Power Production Prediction
Application for predicting PV production and revenue from rooftop solar panel data, using machine learning forecasting models on real-world energy data.
Experience
Designed an uncertainty-aware fusion framework for soil moisture estimation by combining CYGNSS GNSS-R retrievals with a probabilistic SMAP L4 forecaster. Implemented Wasserstein-2 barycenter fusion with adaptive weighting and quality-control gating, improving calibration and accuracy over single-source baselines; paper accepted at IGARSS 2026.
Maintained a 4.6/5 quality rating across 1,000+ reviews in 15+ projects, delivering reliable training data for model development. Drove OCR validation, adversarial prompt testing, and rubric-based QA workflows to surface failure modes and improve labeling consistency.
Prepared and taught recitation sessions, designed exam papers, and graded student work across 5 courses (Data Science, Python, Discrete Mathematics, Probability, Image Analysis); all 5 professors rated me 4/4 in willingness, collaboration, availability, effectiveness, consistency, and initiative.
Tutored 2 undergraduate students per semester across 5 core Mathematics and Programming subjects (Linear Algebra, Python, Discrete Mathematics, Probability, Numerical Analysis); only 1 student did not pass their exams.
Education
Grade: 9.8 / 10. Specialization: Machine Learning, Artificial Intelligence & Signal Processing.
Applied Mathematics, focusing on ML theory. Core courses: Linear Algebra, Machine Learning, Optimization, Python, Statistics, Algorithms, Numerical Analysis.