Project Portfolio

Industrial R&D and Personal Projects

Mixture-of-Experts Ensemble for Skin Lesion Classification Aug 2025

APS360 Course Project | University of Toronto

ARCHITECTURE MoE Gating Network, EfficientNet (B0, B2, B3), MobileNet-V2

PERFORMANCE 82.5% Accuracy, 0.966 AUC, 74.8% Balanced Accuracy

• Developed a Mixture-of-Experts (MoE) inspired ensemble using a top-k gating network to selectively weigh individual CNN experts.
• Mitigated extreme class imbalance in the HAM10000 dataset using weighted cross-entropy loss and automated data augmentation (RandAugment, CutOut).
• Implemented a two-stage training workflow: fine-tuning individual experts followed by frozen-backbone gating network optimization.
• Achieved superior robustness over baseline SVMs, particularly in clinical recall for malignant classes.

ML-Driven Friction Reducer Optimization in Hydraulic Fracturing Oct 2025

National Energy Services Reunited | Presented at CSChE Montreal

• Utilized field data to model wellhead pressure as a function of slurry composition and chemical additives.
• Optimized Friction Reducer (FR) injection profiles to reduce FR consumption while transmitting adequate pressure to fracture the rock.

Hydrometallurgical Extraction & Quantitative Analysis of Copper Ores Dec 2025

CHE204 Course Project | University of Toronto

TECHNIQUES Complexometric Titration, Spectrophotometry (UV-Vis)

ANALYSIS Statistical Error Analysis & Pugh Method Feasibility

• Designed and executed an experimental procedure to quantify copper metal in concentrate samples from two mine sites.
• Analyzed chalcopyrite (CuFeS₂) samples converted to Copper (II) Oxide to determine %Cu by weight.
• Evaluated analytical accuracy between titrimetric and colorimetric methods to replace unreliable ICP instrumentation.
• Formulated procurement recommendations based on experimental yield and logistics/environmental history.