Avatar

Sumith Reddi Baddam

Data Scientist

Amazon Alexa Shopping

Biography

I am working as a Data Scientist in Amazon Alexa. I primarily work on enabling/improving Alexa Speech Recognition (ASR) for Shopping domain. I build machine learning models to improve the ASR for single-turn and multi-turn conversations, perform statistical tests (A/B testing) for launch of new features and perform statistical analysis for identifying customer prone errors and implement solutions for improving the customer experience while having conversations with Alexa realted to Shopping domain.

Prior to joining Amazon Alexa, I previously worked as a Software Development Engineer at Amazon Web Services (AWS) primarily on building end-to-end machine learning applications and pipelines for AWS CloudFormation feature to estimate resource provision time for customers.

I completed two master’s degrees in Data Science from Indiana University Bloomington and IIIT Bangalore. Prior to this, I worked as a Data Scientist at Cisco Systems India for 2.5 years where my key focus was on building large-scale machine learning models to improve the quality of Cisco’s products, internal workflows and productivity of engineers. While working at Cisco, I have filed a patent for building a customer centric bug prioritization feature to enable faster resolution for issues faced by customers and have published 2 papers related to predictive modeling using Deep Learning algorithms.

I was also a key note speaker at International Conference of Business Analytics and Intelligence at Indian Institute of Management Bangalore in 2017 and at Indian Institute of Science in 2018.

Interests

  • Deep Learning
  • Natural Language Processing
  • Machine Learning
  • Predictive modeling
  • Statistics

Education

  • Master of Science in Data Science, 2020

    Indiana University Bloomington

  • M.Tech in Information Technology, 2017

    International Institute of Information Technology Bangalore

  • B.Tech in Information Technology, 2017

    International Institute of Information Technology Bangalore

Skills

Programming Languages

Python, R, Java, C++, C, MATLAB

Machine Learning

TensorFlow, Keras, OpenCV, AWS SageMaker, Scikit-learn, Tableau

Security

Software Security, WebApp Security

Web development

Django, Flask, React, AngularJS, Javascript, HTML

Database

AWS, SQL, MongoDB, Google Cloud, JDBC, NoSQL, ZoDB

Cloud Platforms

Amazon Web Services (AWS), Google Cloud Platform (GCP)

Experience

 
 
 
 
 

Data Scientist

Amazon Alexa Shopping

Apr 2021 – Present Seattle, United States
I primarily work on enabling/improving Alexa Speech Recognition (ASR) for incoming traffic related to Shopping domain on Alexa for multiple locales. I build machine learning models to improve the ASR for single-turn and multi-turn conversations, perform statistical tests (A/B testing) for launch of new features and perform statistical analysis for identifying customer prone errors and implement solutions for improving the customer experience while having conversations with Alexa.
 
 
 
 
 

Software Development Engineer (ML)

Amazon Web Services (AWS)

Jun 2020 – Apr 2021 Seattle, United States
Worked on building machine learning applications and pipelines for AWS CloudFormation service. I built an end-to-end machine learning application/feature to estimate the resource provision time for deploying the cloud infrastructure on AWS. The pipeline consists of weekly jobs for data extraction from S3 buckets, pre-processing using lambda functions and the prediction models built and hosted on Amazon SageMaker. I also productionized this application into the AWS CloudFormation service’s workflow.
 
 
 
 
 

Data Scientist

Cisco Systems

Jan 2017 – Aug 2019 Bengaluru, India
Worked on building machine learning models to improve the quality of Cisco products and its internal workflow:

  • Recommendation engine for identifying peer reviewers for testing on Cisco’s code review platform using NLP.
  • Keywords extraction and document classification of service request cases using unsupervised LDA modeling.
  • Classification of Cisco products into various categories to help the sales teams improve their revenue generation.
  • Identification of files that get impacted when set of files are committed to repository using Association Mining.
  • Clustering the features of products based on the text data and summary fields with NLP and K-means clustering.
  • Software upgrade recommendations to customers using random forest and data mining.
 
 
 
 
 

Data Semantics Intern

DataWeave Software Pvt. Ltd.

May 2016 – Jul 2016 Bengaluru, India
I implemented an algorithm that performs clustering of the products from various e-commerce websites and provides pricing insights to our customers. This product was built to scale to 10 Million concurrent users using distributed scheduling of jobs. I also built an automation engine that classifies the products into various categories using SVM, random forest and neural networks. I managed to improve the accuracy of the classification models from 81% to 90%.
 
 
 
 
 

Big Data Analytics Intern

Zettamine Labs (Apple Inc. client)

May 2015 – Jul 2015 Hyderabad, India
I built an end-to-end product that performs web-scrapping of data, analyse the customer reviews and provides insights to the manufacturers (Apple Inc.). These insights range from “what issues are the customers facing in iPad?” to “What extent is battery drain issue affecting the customers?” My research publication on building this product using Natural Language Processing was selected at the MongoDB Conference - New York, 2015.

Patent & Publications

NeuralCook – Image2Ingredients and cooking recommendation using Deep Learning

IUB 2020 May 2020
Deep learning application to identify ingredients from cooking dishes images and recommend dishes to cook, given a set of ingredients. This application leverages NLP and Computer Vision to learn semantic knowledge using joint embeddings.

Customer Success using Deep Learning - Patent

Cisco Systems Inc. Nov 2018
Built a prediction model for prioritizing the bugs identified during testing phases whether to be fixed fast or can wait. Unstructured bug attributes like descriptions, error log files along with 170 structured fields were used for building the system. It was implemented using LSTM and CNN in Keras and TensorFlow.

Intelligent defect creation system using Siamese CNN LSTM techniques

ICBAI 2018 Nov 2018
Implemented a duplicate bug detector that identifies whether a newly created bug is a duplicate of an existing bug in the Cisco Defect Tracking System and then retrieves all similar bugs from the database with an accuracy close to 90%.

Prediction of issues customers face in a software using unsupervised learning

ICBAI 2017 Dec 2017
Implemented Deep Neural Network model in TensorFlow which predicts the issues customers might face in a Cisco product post its release, helping developer teams fix them prior with an accuracy of 95% on Cisco’s Next-Gen devices.

Projects

NeuralCook – Image2Ingredients and cooking recommendation using Deep Learning

Indiana University Bloomington Mar 2020
Deep learning application to identify ingredients from cooking dishes images and recommend dishes to cook, given a set of ingredients. This application leverages NLP and Computer Vision to learn semantic knowledge using joint embeddings.

Human Computer Interaction

Siemens India Oct 2018
Virtual agent that acts as a receptionist. A 3 layered architecture that has dialogue management, video analysis, speech to text and text to speech models. Built deep learning model for dialogue management. Video analysis involved fate detection and recognition. Speech synthesis was using API.

Object Recognition using Deep Neural Networks

IIIT-B Mar 2018
Visual categorization of objects using Convolution Neural Networks in Python.

Automated Essay Grading System

IIIT-B Mar 2018
The students in an interview were asked to write essays on specific topics and the task is to grade those essays. I have built a POS Tagger module using SCRDR algorithm and used this as a feature along with other word features. The model was trained using neural networks.

Visual Categorization with Bags of Key-points

IIIT-B Mar 2018
Classification of objects in an image using SIFT descriptor and Support Vector Machine classifier. Implemented the paper by Xerox Research institute.

Data Analytics on Karnataka State Government Education data

IIIT-B Mar 2018
Association rule mining, classification, clustering and statistical analysis on the Karnataka state secondary high school dataset to find the insights and suggest the government to function better.

Carpooling Web Application in NodeJS with Object Oriented Programming paradigm

IIIT-B Jul 2017 – Dec 2017
Built a web application using crowd sourcing where people traveling for same destination can share a ride saving money, fuel and pollution. Django web framework was used for building the application.

Smart Canteen System

IIIT-B Jul 2017 – Dec 2017
The project was implemented for our hostel canteen system. It aims at reducing the waiting time of students and professors in the queue by analyzing the queue length using image processing and estimates the time to be taken by using the previous historical data.

Cloud data analytics for efficient usage of water for a smart city initiative

Broadcom Hackathon Jul 2017 – Dec 2017
Storage of data regarding the environmental conditions of the plants along the road dividers using Broadcom WICED sensor and doing an analysis on the data to predict the amount of water usage required for the plants and the time at which they need to be watered.

Object Graph Database

IIIT-B Jul 2017 – Dec 2017
Building an object graph for cricket using Object Oriented Database (OODB) model using JDBC and MySQL storage and Spring web framework. The application was hosted on IBM Bluemix.

App Store using Object Oriented Programming in Java

IIIT-B Jul 2017 – Dec 2017
Programmed database for online app store using Java, JDBC and MySql. A model similar to google play-store.

Object Oriented Programming in C++

IIIT-B Jul 2017 – Dec 2017
Implemented a full-fledged Game of Scrabble in Java making use of Object Oriented Programming.

Contact