Abdul Hannan Kanji

Applied Data Scientist, Sway AI

About Me

Hi, I am Abdul Hannan, an Applied Data Scientist at Sway AI. My research interests lies in Fairness and Safety in Machine Learning.

Previously, I worked on maintaining the COVID-19 Forecast hub under Prof. Nicholas Reich.

When not in front of my laptop, I usually hike or go for a run. I also play the drums, mostly as a way to de-stress!

I recently created a bot to ask for any available resources at this ongoing deadly second wave in India.

Projects

Seldonian Library

Project homepage

This is a Python library that can be used by developers to train ML models, and RL Agents using the Seldonian Framework explained in this paper. You can use any of the existing optimization techniques in scipy.optimize package or also use the CMA-ES implementation in the library.

Fairlearn

Homepage

Fairlearn is an Open Source python library started at Microsoft that can be used to train ML models under various constraints. I implemented the Equality of Opportunity constraint on Fairlearn and worked on experiments on scalability with respect to training models using Fairlearn.

Experience

Sway AI

Applied Data Scientist

July 2021-present

At Sway AI, I work on various Data Science related tasks and build the necessary infrastructure around the models that we develop within Sway AI.

Reich Lab, UMass Amherst

Research Programmer

June 2020-May 2021

https://covid19forecasthub.org

At Reich Lab, I work on maintaining and developing features in the hub like the visualization, the data acceptance pipeline and the website.

Intuit Inc., Mountain View, CA

Software Development Engineer 2

August 2017-July 2019

https://intuit.com

At Intuit, I worked as a software developer on various Quickbooks products.

Education

University of Massachusetts, Amherst

Master's, Computer Science

2019-2021

https://umass.edu

Coursework

  • Machine Learning.
  • Neural Networks.
  • Probabilistic Graphical Modeling.
  • Advanced Natural Language Processing.
  • Reinforcement Learning.
  • Algorithmic Fairness and Strategic Behavior.
  • Advanced Algorithms

PES University

Bachelor's of Engineering, Computer Science

2011-2015

https://pes.edu

Coursework

  • Data Mining
  • Natural Language Processing
  • Cloud Computing and Big Data

Other Projects

Voice Triggers

Project homepage

Worked on the CMUSphinx library on porting it to mobile devices. This is a Java library that maps speech to a unique identifier. This can be used in a sbtract way to map these IDs to specific actions/triggers.

This speech recognition library is language agnostic. As a part of the project, we also ported some modules of the Sphinx package to Android.

Unified Heterogeneous Hybrid Cloud Management

This Project aims at creating a unified management interface for the vSphere and Openstack clouds which aids in easy interoperability between the two clouds and flexibility of choice to deploy applications in either one of the infrastructures.

In order to facilitate the above requirement, a plugin for VMware vSphere was developed which could seamlessly interact with the OpenStack Deployment and perform all the basic cloud functionalities like Instance deployment, suspension, etc. with a single Unified platform without gaining the extra knowledge of usage of the OpenStack interface. Additionally, a seamless private networking interface was also created between the VMs of a tenant which are distributed between the two clouds.

Caching for Distributed and Hybrid Clouds

Paper

Hybrid clouds help in establishing a cost effective, reliable and scalable solution to limitations in traditional on-premise data-center. There is an increasing requirement for federated hybrid cloud solutions to allow private clouds to leverage different public cloud providers. Federated cloud solutions should accommodate infrastructure constraints, balance load between different clouds and keep the cost at a minimal for the client. The following paper proposes an architecture for hybrid clouds called Simple Cloud Federation (SCF). It differs from existing architectures in that it leverages the hierarchical organization features already found in existing cloud systems. It is thus general and easy to implement across a variety of cloud systems, including OpenStack and Amazon. The paper contains details of our OpenStack implementation.