Rudra Mangukiya

Architected and optimized a production-level machine learning pipeline for computer vision, integrating a custom Multihead Regressor model designed for multi-output weight estimation across diverse dish classes. Extensive experimentation across neural architectures, training strategies, augmentation pipelines, and hyperparameters—achieving significant accuracy and robustness improvements. Built a fully reproducible ML experimentation workflow using MLflow (for experiment tracking) and DVC (for dataset versioning), ensuring transparent, research-grade model development. Used AWS GPU infrastructure to accelerate large-scale deep learning training, reducing iteration cycles and enabling rapid experimentation. Developed advanced evaluation Pipeline including automated metric reports, detailed performance visualizations, and Grad-CAM interpretability pipelines for model insight and debugging. Conducted deep dataset analytics using FiftyOne, identifying distribution issues, annotation inconsistencies, and failure patterns to guide targeted data improvements. Utilized Neutron for architectural visualization and debugging, ensuring correctness and clarity of complex CNN models. Worked within an agile engineering environment using Git, Jira, and CI/CD-aligned ML practices to maintain clean development workflows and deliver consistent sprint outcomes. Documented model workflows, experiment findings, and technical insights on Confluence, shaping both engineering decision-making and data collection strategies. Gained strong expert-level intuition in how architectural changes, training configurations, and dataset properties directly impact computer vision model performance in real-world applications. 👉 A special thanks to Florian Reif and Bharath Ramachandraiah for their constant support and guidance during this journey.

Built an end-to-end RAG-based JIRA ticket classification system using LLMs, integrating APIs, databases, and AWS infrastructure. Automated verification and testing workflows, improving efficiency and reducing manual effort. Developed a simple user interface to demonstrate and support solution adoption. Contributed to data preprocessing, prompt engineering, and model validation to enhance performance. Collaborated with cross-functional teams to integrate AI solutions into existing workflows. Supported scalability, containerization, and CI/CD deployment for reliable production use.

• Developed user-friendly interfaces for websites and apps at Sarang Infotech. • Conducted data analysis to extract valuable insights for project enhancement. • Contributed to innovative digital solutions with a keen eye for detail.