Ayush Kumar

Engineering a GPS-based track labeling system to identify train routes, improving the data ingestion pipeline to compare 10M+ data points across 200+ railway lines. Developed and validated Pydantic data models for processing and storing large-scale GPS data from a RESTful API, integrating Python, Bazel, and SQLite for efficient data handling. Built train route visualizer using the Folium library, with automated testing through UnitTest and Pytest, and version control implemented via GIT and GitHub Actions.

Implemented deep neural networks for adversarial patch regeneration, achieving a 96% face detection rate on the CelebA dataset. Worked on ML/DL models security and resilience using Domain Separation Networks, achieving 92% image classification accuracy on the MNIST dataset. Conducted R&D and experiments on various datasets performing attack simulations and optimizing code for robust models. Collaborated with research scientists on adversarial machine learning, contributing to two scientific publications.

Led a team of 3 engineers to develop scalable RPA and AI/ML solutions using agile methods, achieving 50% faster data processing. Designed and implemented test-driven development and code reviews addressing customer problems to automate over 10 software applications. Built a machine learning model using OCR APIs for invoice parsing, achieving 80% accuracy, and deployed it on on-premises infrastructure. Engaged with customers and stakeholders to translate business challenges into technical solutions.

Grade: 1.5/5

Grade: 1.3/5 (German Equivalent)