Close

Manjunath Bhat

Deep Learning and Computer Vision Enthusiast

Download Resume

Who is Manjunath Bhat?

Manjunath Bhat is a third-year undergraduate student from the Department of Mechanical Engineering at Indian Institute of Technology(IIT), Kharagpur. He transforms into "The Bhatman" when he is coding and trying to solve problems. He is interested in Deep Learning, Computer Vision, Robotics, and path planning algorithms. The Bhatman has always been fascinated by mathematics and physics, ever since he was a kid.

He is an active member of the coding team of Kharagpur RoboSoccer Students' Group(KRSSG). In this group, he is part of the team that takes part in the International RoboCup Small Sized League(SSL) held annually.

Education

Indian Institute of Technology, Kharagpur

July 2017 - June 2021(expected)

Bachelor of Technology(Hons) in Mechanical Engineering

Secured a CGPA of 8.77/10 (After six semesters)

SKCH Composite Pre-University College

June 2015 - March 2017

Class XI and Class XII

Prarthana Central School

2015

Class X

Experience

Goldman Sachs

Summer Analyst, Securities Division

Project: Internalization Attribution across stable business sources and providing Relative Value Trades Functionality. Contributed to the Index Pricing Tool by processing the output of the Linear Optimizer used in the tool, and attributing the Internalization amount across different business sources, by prioritizing them according to their stability. Provided functionality to price multiple new trades (Relative Value Trades) on indices.

Projects

Google Summer of Code 2019 at FluxML (The Julia Language)

Enriching FluxML's model zoo with deep learning models : Spatial Transformer Network, VAEGAN, EBGAN, StarGAN and GRCNN. I also contributed to the Flux library by adding new dropout, normalization layers, wrappers for convolutions and working on the backend integration of an Automatic Differentiation package (called Zygote) with the Flux Library.

View Project on Github

Random Sampling Path Planning Algorithms used in Robotics

In highly dynamic and adversarial environments, random sampling algorithms such as Rapidly-exploring Random Trees(RRT), RRT-connect, RRT-Star, RRT-Star with Artificial Potential Field are suitable for fast computation of path. I have studied and implemented these algorithms in 2D for any given obstacle field.

View Project on Github

Finite State Machine Codebase for RoboCup SSL

This is the main codebase based on the Finite State Machine(FSM) architecture focused on the performance of KRSSG in RoboCup SSL. The data flow and communication of commands takes place through ROS. The OMPL library is used for robust path planning. Roles and basic skills required in Robosoccer such as "Go to ball", "Kick to point", "Goalie" are implemented in the FSM structure.

View Project on Github

ConnectAll - Enabling the differently abled.

Developed a web app that bridges the communication gap that exists among deaf, blind and mute people. The app provides a chat and call platform, that converts the speaker’s voice to text in real-time, so that a deaf person can understand and respond. It also enables blind people to respond to text messages by converting text to an automated voice. Real-time note making, when notes are being dictated is another feature of the app. The app also provides a feature for personalized book narration. These features have been automated with a Zulip chatbot that responds on the Zulip Chat platform when pinged with a request.

View Project on Github

MedAI

It includes an interactive symptom checker, which works with apriori clustering and naive bayes styled algorithms. A feature of physical doctor prescription parser is also included which works with hadwritten-OCR and medical semantic recognition to extract names of drugs and their dosage and creates a Google Calendar reminder to ensure the user takes the medication at the appropriate time.

View Project on Github

Maze solving Robot

A three-wheeled robot with two driving wheels and one castor wheel for steering, that can find the shortest path between source and destination in a maze, and can follow the path generated. Various techniques of Image Processing such as Edge Detection, Contour Detection, and Hough Transforms were used. Dijkstra's algorithm was used to generate the shortest path.

View Project on Github

Skills

Get in Touch