About me

“Life is like a chaotic system, where you might need to find another basin of attractor to keep the system in equilibrium.”

I’m Kunal Rathore — an Applied AI Scientist with PhD in Environmental Sciences from Oregon State University, my research work comprises of developing hybrid modeling techniques including but not limited to hydrology, water-quality simulations, remote sensing, and Deep Learning. With hands-on industry experience building and deploying production ML systems at Seagate Technology and Persistent Systems, I bring a strong track record of translating research into scalable, industry-ready solutions.

My work spans Generative AI & LLMs, Scientific Machine Learning, Explainable AI, Computer Vision, and Data-driven Predictive Modeling from enterprise NL-to-SQL pipelines and real-time sentiment platforms to digital twin models for environmental forecasting.

Currently, I research with Prof. James Watson at the Socio-Environmental Analysis (SEA) Lab, developing hybrid AI models that combine mechanistic science with deep learning for ecosystem prediction and critical transition detection.

Featured Projects — View All →

• NL2SQL: LLM-Powered Natural Language to Database Query Engine — Multi-stage prompt-engineered pipeline built at Seagate Technology that converts plain-English questions into optimized SQL queries at 91% accuracy, cutting query fulfillment from days to <3 seconds.
• CatRisk: Probabilistic Catastrophe Risk Model for Hurricane Loss Quantification — An enterprise-scale probabilistic catastrophe model that turns 170 years of NOAA storm data and US economic data into a 10,000-year hurricane loss simulation — pricing insurance risk down to the individual property.
• Hybrid HAB: Digital Twins for Predicting Harmful Algal Blooms — Hybrid ML + mechanistic modeling framework for real-time bloom prediction and “what-if” scenario testing in lake ecosystems. Funded by U.S. Army Corps of Engineers.
• C-Star: Predicting Unresolved Variables in Dynamical Systems — Universal Dynamic Equations (UDEs) that uncover hidden drivers of ecosystem regime shifts from observable data alone — delivering superior early-warning capability with less data than conventional methods.

• GEPC: Making AI Decisions Understandable — Scalable Explainable AI framework achieving 84% label-transfer accuracy across 158 ImageNet categories, reducing annotation effort by 85%. Published at IJCAI 2024 XAI Workshop.

• If our interests match, I am always open to collaborating and starting something adventurous! Please feel free to connect.