Projects

Generated image:

Facial Key Points Detection

Tools: Pytorch, Seaborn

• A CNN was defined and trained on Youtube Faces Dataset to identify 68 facial key-points.

• Using Haar cascades, face was detected.

• Edge detection (using Canny detector), HOG, BRIEF, FAST algorithms were explored.

Image captioning

Tools: Pytorch

• A pretrained ResNet extended with untrained linear layers was trained on the MS COCO dataset.

• YOLO, Attention encoder techniques were employed to encode the features of the input image.

• An RNN (with LSTMs) was employed as a decoder to generate highly accurate captions on new input images

Fruit Or Not

Tools: Matlab, Arduino C

• Otsu thresholding was used to segment the images

• Arduino was mounted on a drone to analyse the ripeness of areca nuts, as these are bore by trees which are on average as tall as 20m.

FishBot

Tools: EagleCAD, EDA

• A bouyant robot designed to bio-mimic fishes.

• Designed and fabricated integrated circuit boards:

  • For power supply

  • Control of the fish

  • RF Controller

• Semi Finalists in E-Yantra.

Wall E

Tools: Raspberry Pi

• Life size replica of Pixar's Wall-E robot.

• Built the robot.

• Designed and fabricated integrated circuit board for controlling locomotion of the robot.

MLJBase

• This package implements core functionality for the MLJ machine learning framework.

• I contributed multi-class statistical metrics.

• Post reorganization, this code is moved to StatisticalMeasures.jl

ModelingToolkit

It is an acausal modeling framework.

• Was instrumental in adding a new and convenient domain specific language (DSL) to define models in Julia.

Checkout the repo.

NgSpice

It is a Julia wrapper for Ngspice simulator. It is an opensource spice simulator for electric and electronic circuits

JuliaSim

It is a next-gen cloud based modeling platform by JuliaHub. I work on internal tooling for building and deploying the app and I work on building out the modeling standard library (more on it is here)