Divyanshu Dubey

AI/ML(Machine Learning) Researcher

MicroAcquire • Part time

Jan 2024 - Nov 2025 • 1 yr 10 mos

Project Overview: Collaborated with the Micro-Nano R&D Centre to apply cutting-edge machine learning and data science methodologies to high-dimensional scientific datasets for multiple CSIR-backed and academic research projects. The Challenges: Scientific datasets are notoriously complex, non-linear, and high-dimensional. Traditional modeling falls short, requiring advanced, physics-informed AI architectures, hyper-precise validation frameworks, and deep model interpretability to make the predictions viable for academic publication. The Solution & Execution: I took charge of the core ML infrastructure and predictive modeling:Advanced Architecture: Built and deployed deep learning and ensemble pipelines using GBRNN and LSTM models. Hyper-Optimization: Integrated Bayesian hyperparameter tuning and enforced strict 5-fold and 10-fold cross-validation to ensure zero overfitting. Explainable AI (XAI): Applied SHAP (SHapley Additive exPlanations) to interpret complex model decisions, translating black-box AI into actionable scientific insights. The Results: Successfully bridged the gap between computational physics and artificial intelligence. The work resulted in an independent research publication where our ensemble models achieved a remarkable predictive accuracy of R^2 == 0.97+ for material informatics.