experience

I work on a problem at the intersection of geophysical science and applied machine learning: how do you build models that reliably detect and forecast rare, extreme events from high-dimensional, noisy observational data — when the underlying physics is partially understood but too complex to model analytically?

My approach is physics-first. Before modelling, I study the documented physical mechanisms behind marine heatwave formation — mixed layer dynamics, air-sea heat flux anomalies, oceanic circulation patterns. That understanding shapes every modelling decision: which features to engineer, what architecture makes physical sense, where to constrain the model, and how to interpret what it learns.

• Designing and evaluating deep learning architectures (ConvLSTM, temporal CNNs) for spatiotemporal sequence modelling on gridded ocean reanalysis data
• Statistical characterisation of extreme event distributions — understanding tail behaviour before attempting prediction
• Physics-informed feature engineering: encoding domain knowledge as model inputs rather than expecting the network to discover physical relationships from scratch
• Evaluating model outputs against known physical drivers to ensure predictions are physically consistent, not just statistically accurate

Worked within a 50-person team building ML-driven solutions for e-commerce and retail clients — from model development through to business impact measurement.

• Developed and optimised deep learning models (TensorFlow, PyTorch) for product price optimisation and customer segmentation, contributing to a 15% increase in sales for e-commerce clients
• Built data analytics pipelines using Tableau and Apache Superset to extract actionable insights from large retail datasets, improving sales forecast accuracy by 10%
• Collaborated on model evaluation, feature engineering, and deployment workflows alongside senior engineers

Four years of hands-on research in the Materials Energy Manufacturing Sustainability (MEMS) Lab. My thesis focused on friction stir welding (FSW) of dissimilar materials — AA6082-T6 aluminium alloy and DP780 dual-phase steel — targeting lightweight automotive applications.

• Designed and executed FSW process parameter studies; characterised joints using HR-SEM, EDS, XRD, EBSD, nano-indentation, and UTM mechanical testing
• Correlated intermetallic compound (IMC) formation at the weld interface with joint failure mechanisms and mechanical performance under varying process parameters
• Developed ML models (Random Forest, SVM, linear regression) to predict UTS from process parameters — connecting data-driven methods to physical understanding
• Presented at THERMEC'23 (University of Technology Vienna, Austria); secured 3rd place at IIT Madras national symposium
• Teaching Assistant for MM2010 (Physical Metallurgy) and NOC24-MM14 (NDT) — led weekly lab sessions for a batch of 70 students

1st-author manuscript under review — Journal of Manufacturing Processes.

Alongside research, I contributed to the broader IIT Madras student community through structured roles in career development and departmental engagement.

• Active member of CDC-R (Career Development Cell – Research) — supporting research scholars with career planning, industry connections, and professional development
• Participated in departmental placement cell activities, connecting students with research and industry opportunities

Conducted technical and financial valuations of large-scale plant and machinery projects (INR 150–200 Cr range), providing assessments that supported strategic procurement and investment decisions across manufacturing and infrastructure sectors. Developed rigorous cross-functional reporting skills working across engineering and finance teams.

Managed a production team in steel bridge fabrication. Optimised machining parameters through data-driven analysis of production records, contributing to a 9–10% improvement in production throughput. First hands-on exposure to manufacturing process data and the value of systematic parameter analysis.