Projects Portfolio
This site showcases my research and projects at the crossroads of chemical engineering and artificial intelligence. My PhD focuses on creating physics-informed machine learning models for reactor modeling and process optimization in energy and environmental sectors. I've applied techniques like Neural ODEs and deep learning to various challenges, including crude oil characterization and lake temperature prediction. With expertise in process systems engineering and AI interpretability, I seek to connect foundational science with modern machine learning. Please explore my projects and reach out for discussions or collaborations.
Chemometrics of Crude Oil from Spectral Data
PhD Research
This project focuses on developing machine learning models to predict physicochemical properties of crude oils using FTIR spectral data. Leveraging 1D Convolutional Neural Networks (1D-CNNs), I aimed to extract meaningful patterns from complex spectral signatures to estimate properties such as viscosity, density, and sulfur content. The approach significantly reduced prediction errors compared to conventional methods and emphasized model interpretability using techniques such as saliency mapping and SHAP values.
Physics-Informed and Interpretable Modelling of Hydrocracking Kinetics
PhD Research
In this study, I integrated domain knowledge of chemical kinetics with neural ordinary differential equations (Neural ODEs) to model hydrocracking reactions in petroleum refining. The model was trained on synthetic and industrial data, capturing non-linear dynamics and temperature-dependent behaviors of multi-lump reaction networks. By incorporating Arrhenius kinetics into the training process, the model achieved both physical consistency and high predictive accuracy.
Physics Informed Modelling of Great Lakes Water Surface Temperature
Research Project
This ongoing project explores the use of Long Short-Term Memory (LSTM) networks combined with physics-based constraints to predict lake surface temperatures across buoy locations in the Great Lakes. The model leverages historical meteorological and hydrological inputs to learn temporal dynamics, while physics-informed loss functions ensure consistency with thermodynamic principles.
Surrogate Modeling for Real-Time Optimization of Hydroprocessing Units
Research Project
This ongoing project focuses on building AI-based surrogate models for plant-scale hydroprocessing units, including hydrocrackers and hydrotreaters. Using real industrial datasets, I trained models to approximate complex plant behavior under varying operating conditions. The models are designed to aid in optimization and predictive control applications, enabling faster decision-making and improved energy efficiency.
ChemSimGuide: AI-Powered Chemical Simulation Assistant
Google GenAI Capstone Project – Honorable Mention
This project addresses the inherent complexity and usability challenges faced by chemical engineers and researchers when setting up simulations with advanced computational tools like Cantera. ChemSimGuide is an AI-driven conversational assistant designed to simplify chemical simulation setup through intuitive, natural language interactions.
Leveraging Google's Gemini-2.5-Pro language model, ChemSimGuide translates user-described simulation goals into executable Python scripts, automatically integrating detailed knowledge from Cantera’s extensive documentation using Retrieval-Augmented Generation (RAG). The conversational interface, powered by LangGraph, systematically guides users through clarifying simulation parameters, making model selections, and troubleshooting common simulation errors.
