Kaustav Deka

Garyfallidis Research Group, Indiana University · Bloomington, IN

• Accelerated EuDX tractography by 12×, reducing runtime from hours to minutes via OpenMP parallelization and Cython optimization for large diffusion MRI datasets
• Built an end-to-end diffusion MRI pipeline using 15+ DIPY methods and a Bash CLI, automating 20+ steps previously executed manually
• Delivered a PyQt GUI that reduced DIPY setup complexity by 60%, enabling non-technical users to run advanced tractography workflows
• Developing the GLIDE uncertainty-adaptive tractography algorithm within DIPY's Cython propagator architecture; evaluated on HCP data with RecoBundles across 80 bundles
• Contributed to the FORCE microstructure estimation pipeline (under review at Nature Methods), designing stability experiments including library resampling and input perturbation
• Created publication-quality scientific figures: Nature-style SVG pipeline diagrams, multi-panel matplotlib comparison figures, and Blender-rendered white matter visualizations

DIPY Neuroimaging Library · Remote

• Shipped 50+ open-source contributions to a Python library with 10,000+ users, including performance improvements cutting memory use by 30%
• Implemented lazy loading to decrease import time by 45% and reduce initial memory footprint from 250 MB to 140 MB
• Improved documentation usability by refactoring a website serving 5,000+ monthly visitors, reducing developer onboarding time by 25%

RGBSI · Bengaluru, India

• Improved product adoption by 20% by building Power BI analytics dashboards to guide data-driven feature prioritization for a SaaS platform
• Delivered 4 enterprise software implementations for 200+ end users on schedule, achieving 95%+ stakeholder satisfaction

RGBSI · Bengaluru, India

• Built 25+ Power BI and Business Objects dashboards analyzing historical trends, driving 3 major system enhancements and 15% efficiency gains
• Reduced production support tickets by 35% through proactive analysis and system gap identification

FORCE: FORward modeling for Complex microstructure Estimation

Co-author · Under review at Nature Methods

Novel forward modeling approach for estimating brain tissue microstructure from diffusion MRI data. Contributed stability experiments, figure design, and pipeline validation.

Cross-Modality Brain MRI Synthesis: A Comparative Study

Co-first author · Indiana University Bloomington

Comparative study of five deep learning architectures for bidirectional T1↔T2 brain MRI synthesis with multi-metric evaluation and robustness analysis.

Indiana University, Bloomington — GPA: 3.85 / 4.00

Machine Learning · Deep Learning · Computer Vision · Optimization · Information Visualization

Bangalore Institute of Technology, Bangalore, India