My name is Abhishank and I Craft Products!
I could waste your time by convincing you that I always " Accomplish X by doing Y leading to Z ", but ultimately, my skills are as temporary as every language I know and as fickle as every tool I use. As temporary as things that will be useless in 10 years. As fickle as things that can be learned by anyone.
The things that LAST are not complete, they are in PROGRESS. So what I will tell you is that I AM IN PROGRESS. I hold on to my methods loosely, embrace flexibility and modularity, am willing to learn, expect to be wrong, and am confident when I am right.
My various internships have given me a chance to dip my toes in a variety of fields. I became a Lean Startup Evangelizer at 1 checkpoint, a Deep Learning Engineer at another, and a Software Connoisseur at the 3rd. My experiences morphed me into a Tech Lover and a Big Data Enthusiast. Feel free to check out what I've been up to by clicking the tabs above.
As a Product Manager:
As a Product Manager:
As a Software Lead:
This is a research paper I worked on over the past year with 1 other author. We are actively iterating on the paper's implementation of detecting defects in sewer pipes, in order to get it published. The abstract is given below.
A deep multi-class convolutional neural network is built to detect critical points in pipes; specifically joints, connections, and manholes. An accuracy of 91.7% is achieved using a dataset composed of 7 pipe videos. Three hyperparameters are varied: learning rate, batch size, and class weights assess their impact on training. In sum, the high accuracy implies that the model may be overfitted to the given dataset, despite the data augmentations, 5 fold validations, and dropouts used.
I, along with 1 other partner, determined person's current location using an Extended Kalman Filter. I, then, calculated the shortest path to their final location using the A* algorithm (variant of Dijkstra's Algorithm).
View Project
I identified different tissue patterns from different primary sites (lung, kidney, and ovary) by building a convolutional neural network using a Keras Sequential Model (Machine Learning), leading to an accuracy of 80.20% on unseen test data.
Implementation details are provided in the pdf file located in the github repository. File name: "Implementation_Explanation.pdf"
This was the first machine learning personal project I ever tried. I completed this project over the summer of 2018.
I Calculated statistical features (mean, standard deviation, etc) by filtering, signal processing, and spectrally analyzing IMU data from 6 subjects.
I then built a MICD classifier (Machine Learning) to identify whether person is walking, running, sitting, or cycling.