Aakash Maroti

Led the development and implementation of advanced computer vision algorithms for robotic systems, also integrating sensors with robotic arms using ROS. Involved in formulating and managing research proposals, ensuring successful project acquisitions and execution. Presented research insights at conferences. Developed vision systems for inspecting and handling objects, automated systems for agricultural tasks, and enhanced recycling processes through automation. Implemented frameworks for automatic fruit harvesting using reinforcement learning, showcasing expertise in robotics and computer vision.

Using a wide range of techniques from classic image processing to modern state-of-the-art computer vision to innovate and solve problems. Redesigned the computer vision stack to introduce multi-threading, increase code re-usability, enable better testing, simplify update cycles, and bring about significant performance improvements. Leveraged computer vision to automate the installation of apps, selection of profiles, and launching of content across different devices. Patented a new image classification technique that reduced device detection time by 80%. Worked on other projects including automatic crawling and launching of DVR content, detecting and replacing advertisements on live TV, and detecting the type of keyboard present on the screen for automatic text entry.

Performance and Log monitoring system with web-based front end. RESTful Spring mvc webapp, to assist in automation and dynamic deployment.

Android app for easy data collection.

Android app to manage fuel orders from depot.

A RL approach to the classic game of Brainvita, that does not involve any search or look ahead. Also includes the environment.

An approach to solve the classic OpenAI problem of cartpole, with focus on using a more sophisticated exploration strategy. Preprint.

Using NeRF to create a 3d representation of a scene from a single video of the scene. Also generating new video that navigates the same scene from a different perspective.

Set of simple OpenAI Gym compatible environment to assist in bug free developement of Deep Reinforcement Learning agent.

Reinforcement Learning Projects that were part of Deep Learning Nanodegree by udacity. Using various basic Reinforcement Learning algorthims to solve basic environments.

Use CNNs to restore colour in washed out images.

Basic implementation of Capsule networks with dynamic routing between capsules in PyTorch on MNIST.

A fast and optimized implementation of stroke width transform by Microsoft in C++

A complete end-to-end conversational agent to negotiate the best price for purchasing laptops.

Implement actor-critic algorithm from scratch to train an agent for continuous control in the LunarLander environment.

Using Distilbert to classify tweet responses into appropriate categories.

Scrape the web to get information on Grammy nominated artists, their lyrics, and their portrayal in media. Analyse the information obtained to gain insights.

Use knowledge of motion to significantly improve performance of transporter networks in environments with egocentric vision, such as autonomous navigation with sensors mounted on vehicle.

Windows application to control playback on VLC media player with gestures. Camera or on-board webcam is used as a source of input.

Android application that fetches details of a registered Indian vehicle when a number or number plate image is presented.

Using various data mining techniques to predict the placement performance of a final year B-Tech student in MIT Manipal.

Fast lightweight and surprisingly robust approach to classifying certain type of images based on shared static components.

Code and examples of basic deep learning concepts in PyTorch for easy reference.

Preprocess and experiment with the Microsoft search results dataset to increase relevance of search result.

Explored lot of domains with electives like optimization techniques, information retrieval, multimedia communications and distributed systems. Attended workshops like big data analytics, game development in Unity and internet of things by IBM Bluemix. Also stood 2nd in Google Android Wearable Hackathon.

Machine Learning by Stanford University on Coursera. Certificate earned at Sunday, April 15, 2018 6:44 AM GMT. All assignments completed with a final grade of 94.1%

Completed Deep Learning Nanodegree Program along with all of its projects and assignments.

Completed Learning How to Learn with a final grade of 95.1%

Fundamentals of Reinforcement Learning by University of Alberta & Alberta Machine Intelligence Institute on Coursera. Certificate earned at Monday, August 12, 2019 1:40 PM GMT. All assignments completed with a final grade of 100%

CGPA 8.192. Completed courses on Autonomous Systems, Probabilistic Models, Machine Learning, Web Intelligence, Natural Language Interaction, Computational Semantics, Reinforcement Learning, and Thesis on adapting transporter networks for autonomous driving, by using motion information to change point of view.