Hello! 👋

My name is Madhur, and I am an Artificial Intelligence (AI) Developer. I have a master's degree in AI from the University of Groningen.

Priviously, I have worked as an AI/ML Engineer and a student researcher, delving into Machine Learning, Computer Vision, Distributed Systems, and Robotics. I enjoy exploring new technologies that enhance our lives. If you want to know more about me, click here.

To check out my projects, just keep scrolling.


Research Projects

Dual Arm Manipulation of Objects using Deep Reinforcement Learning
  • Master's thesis focused on utilizing Deep Reinforcement Learning (DRL) for object manipulation in a human-inspired robotic system. The research explored the impact of collaborative dual-arm robotics on the success rate and efficiency of grasping unfamiliar objects in novel environments.
  • Technology stack: Python, PyTorch, OpenAI Gym, ROS (Robot Operating System), PyBullet Physics Engine, Gazebo Simulator.
Corporate Carpooling Web Application
  • Developed a robust, fault-tolerant, single-page web application with a fully distributed architecture. The application features asynchronous communication, enabling real-time data processing and seamless interaction across distributed systems.
  • Key features:
    • Fully distributed system with built-in fault tolerance for reliable performance in diverse environments.
    • Asynchronous communication handling for efficient real-time updates.
    • Deployed on OpenStack for cloud scalability, resource efficiency, and redundancy.
  • Technology stack: Vue.js (front-end), Python (back-end), Docker (containerization), Nginx (web server), MongoDB (NoSQL database), Neo4j (graph database), and Ariadne (GraphQL API framework for Python).
From Lego Bricks to Actual Bricks: Style Transfer with CycleGAN
  • Researched and developed a deep learning approach for image-to-image translation using CycleGAN, focusing on unpaired data transformations. The dataset used in the project was meticulously curated and manually annotated to ensure accuracy and diversity in training. This method enables high-quality domain adaptation without paired examples, demonstrating promising results in tasks such as style transfer and object transformation.
  • The project is available on GitHub.
  • Technology stack: TensorFlow (deep learning framework), Python (core programming), Google Colab (cloud-based experimentation and training)
Spatio-temporal Data Indexing
  • Developed a system for ISRO (India Space Research Organisation) to index Spatio-temporal data from satellites and other sources for GIS tools using R-trees, decreasing the retrieval time on test data by 40%. Python and Flask were used for the implementation.
  • Tech stack: Python, R-Trees, AWS.

Experience

  1. Chair of the Programme Committee for AI/CSS at the University of Groningen.
  2. Teaching Assistant at the University of Groningen.
  3. Artificial Intelligence Intern at ABN AMRO Bank.
  4. Student Researcher in High Tech Systems and Materials at Astron.
  5. Artificial Intelligence/ Machine Learning Engineer at Audax Labs.
  6. Student Researcher in Virtual Reality at the Université de Poitiers.