Abhishek Kandoi Senior LLM Eng. at CombineHealth | Computer Science '16 @ IIT

V

Vehicle Trajectory Prediction on NuScenes 3D LiDAR dataset

Apr 2020 - Jun 2020 (3 months)

Trained a Convolutional-LSTM model on the NuScenes 3D lidar and annotation data set to predict future trajectory of target vehicles. Used the NuScenes map expansion pack to generate intermediate representations for trajectory prediction

V

Visual-Inertial SLAM

Feb 2020 - Mar 2020 (2 months)

Developed an Extended Kalman Filter based SLAM algorithm using Stereo Images and IMU data captured from a vehicle (KITTI dataset)

P

Perturbation Resilient Hybrid Walking Robot

Jan 2020 - Mar 2020 (3 months)

Developed a central pattern generator based walking gait for a hybrid wheeled-legged robot. The robot is resilient to leg perturbations and can correct its gait maintaining the specified phase difference b/w its legs.

P

Particle Filter SLAM

Jan 2020 - Feb 2020 (2 months)

Developed a particle filter based SLAM algorithm using Odometry data, laser measurements (lidar) and RGBD images captured from a robot

S

Stop-Sign Detection Using Gaussian-based Probabilistic Learning

Jan 2020 - Feb 2020 (2 months)

Developed a generative Gaussian-based Probabilistic Learning approach for the purpose of detecting stop-signs in RGB images. After segmenting image based on a Gaussian Discriminant Analysis based color model, I used a combination of carefully selected region properties (regionprops) criteria for detecting stop signs.

U

Underwater Acoustic Localization

Oct 2019 - Jan 2020 (4 months)

I'm working on find the direction of arrival of an underwater pinger using an array of hydrophones. I am using the Time Difference of Arrival method with nonlinear least squares estimate and particle filter approach to solve for the pinger's location with respect to our submarine (for the Robosub International competition).

M

Multi-Object Detection using Deep Learning

Sep 2019 - Dec 2019 (4 months)

We developed two multi-object detection methods in this project, YOLO and Faster R-CNN. Both models were implemented with two different pre-trained backbones, VGG16 and ResNet-50, for comparison purposes. With the Faster-RCNN architecture utilizing the VGG16 backend, we achieved a mean Average Precision of 0.7002. We also experimented with photos taken from my smartphone.

F

Fine-grained recognition of North Atlantic Right Whales

Jul 2015 - May 2016 (11 months)

As a team 3 members, we trained a VGG-Net based ML model to classify endangered right whales in aerial photographs. This was a challenging task because the aerial images dataset was small. We did this as part of our B.Tech thesis project, and worked with Torch, numpy, scikit and Matlab to implement our model.

K

Kernel with memory and process management (Project Lead)

Aug 2014 - Nov 2014 (4 months)

We developed a kernel in C (for boot-loader) and C++, with a fat32 file-system and features like memory management, process management and process scheduling. We also implemented basic terminal commands that are available on Linux (including cd, pwd, ls and cat). Some portions of the boot-loader and the Global Descriptor Table were referenced from Bran's Kernel Dev. article.

B

Backdoor Platform

Apr 2014 - Jul 2015 (1 year 4 months)

I created and led the team to build Backdoor, which is a platform for computer hackers to show their talent in a competitive environment. Earlier it was launched within the IIT Roorkee campus, but now it has been made available for anyone over the internet. This initiative led me to found the InfosecIITR group, which is currently ranked number 1 in India (source ctftime.org).

S

SIC/XE Dis-assembler

Jan 2014

Implemented a two pass SIC/XE dis-assembler (no symtab required) in C++, with basic functionality for detecting the different formats of instruction, differentiating between RESB and RESW, incorporating support for jumps and sub-routines.

A

Autonomous Parking Robot

Jan 2014 - Mar 2014 (3 months)

Lead a team of 8 members, as we designed and developed an autonomous parking robot which could navigate through traffic without human involvement. We built the bot using an Arduino board using multiple proximity sensors.

H

Harley: Daily Activity Center for Autistic Children

Created a Kinect application in C# for Microsoft Code.fun.do Hackathon that acts as a daily activity center for Autistic children. Includes several gesture based games to interactively help autistic children improve their motor capabilities daily.

K

Know Your Government

Built a web application using App Engine, Google's cloud platform, that provides information, news, and social updates about political leaders and parties in India. Involved sentiment analysis of the news and social content related to a politician. Selected as a finalist in Google Cloud Developer Challenge 2013, a worldwide competition hosted by Google. Source code: https://github.com/abhikandoi2000/knowyourgov

Abhishek Kandoi

About

Detail

Timeline

Résumé