Shreyas K.

Teaching Assistant @ Delft University of Technology

About

I am deeply passionate about machine learning and deep learning, with a particular focus on the theoretical foundations of machine intelligence algorithms and their analytical study. My past experience primarily involves leveraging neural networks in various domains of astronomy and astrophysics. I have undertaken projects spanning classification and generative models, applying these techniques to solve challenges in computer vision and generative modeling, particularly through GANs. My academic interests extend to analysis, dynamical systems and high dimensional statistics, especially in the context of neural networks. I am intrigued by the theoretical study of overparameterized models and the bounds on their generalisation capabilities. I am also interested in their potential applications in physics and astrophysics.

Country

Netherlands

City

Delft

Industry

Computer Software

Skill

Font Technology, Webpack, Vuex, PySpark, Pandas (Software), Docker, Nginx, Electron.js, Node.js, Software Infrastructure, Data Engineering, TensorFlow, Pytorch lightning, Distributed Computing, PyTorch DistributedDataParallel, Web Development, Ruby on Rails, PyTorch, Data Visualization, Genetic Algorithms

Experience

Teaching Assistant

Delft University of Technology

2025-9 - Present · 1 yr

Delft, South Holland, Netherlands

- Supported MSc courses in Machine & Deep Learning, Statistical Learning, Security & Cryptography, and Constraint Solving. - Led tutorials, created & graded assignments/exams, and assisted students with debugging ML model implementations.

Machine Learning Research Scientist

Relfor Labs Pvt. Ltd.

2022-9 - Present · 4 yrs

Pune, Maharashtra, India

- Built PyTorch Lightning training pipelines on NVIDIA DGX A100 with Optuna-based hyperparameter optimization, accelerating experimentation by ~50%. - Optimized large-scale ETL pipelines, achieving 8× speedup and 30\% storage reduction while processing terabytes of data.

Software Developer

Dalton Maag

2021-11 - 2024-3 · 2 yrs 5 mos

London, England, United Kingdom

- Automated CJK glyph generation using a Python genetic-algorithm POC, demonstrating reduction of production time from months to minutes. - Built and productionized PriceBot (TypeScript, Ruby on Rails), reducing quote generation from hours to seconds. - Designed and deployed containerized application health-tracking with GitLab CI/CD to detect pricing regressions early and prevent potential revenue losses. - Developed glyph and pricing data models for Arabic, Greek, Cyrillic, and Devanagari scripts, expanding non-Latin client coverage. - Applied graph-based optimization in TypeScript to improve project planning accuracy, cutting planning time from hours to minutes. - Contributed to DSedit, an ElectronJS + VueJS internal platform consolidating tooling for type-design workflows.

Machine Learning Engineer

Relfor Labs Pvt. Ltd.

2021-8 - 2021-11 · 4 mos

Pune, Maharashtra, India

- Designed CNN architectures for audio classification in PyTorch achieving ∼98.6% accuracy and ∼0.98 F1-score.

Software Development Intern

FinIQ Consulting

2020-5 - 2020-6 · 2 mos

Nashik, Maharashtra, India

- Developed AngularJS-based Forex trading platform features and built an open-source Python CPU/memory stress-testing module later integrated into cloud testing pipelines.