Sumith Reddi Baddam

Senior Applied Scientist at Amazon Ads, based in Seattle, WA — leading cross-platform advertising integrations across Amazon Sponsored Products and external ad ecosystems.

I lead cross-platform advertising at Amazon Sponsored Products, where my work established Amazon's integration with a major external social advertising platform — building frameworks for real-time bidding, cost-per-click estimation, and incrementality measurement across external advertising ecosystems. I've published research at the Amazon Machine Learning Conference (AMLC) on cross-platform advertising optimization, and have patents pending on dynamic cost-sharing between advertisers and e-commerce platforms in cross-channel advertising.

Before this role, I worked on speech recognition for Alexa's agentic voice-shopping experience, architecting an error-aware reformulation system that improved error correction by 41%. Earlier, I was a Data Scientist at Cisco Systems, building ensemble deep learning models for software defect prediction. I hold master's degrees in Data Science from Indiana University Bloomington and Information Technology from IIIT Bangalore, and serve as a peer reviewer for AMLC 2025 and the Consumer Science Summit 2026.

cross-platform advertising multi-modal ai large language models nlp causal inference bayesian methods

Experience

2023 — now

Senior Applied Scientist

Amazon Ads · Seattle, WA

Lead cross-platform advertising for Amazon Sponsored Products:

  • Pioneered Amazon Sponsored Products' integration with a major external advertising platform, building frameworks for real-time bidding, cost-per-click estimation, and incrementality measurement across external ad ecosystems
  • Invented a real-time cost-per-click estimation algorithm reconciling Amazon's real-time pricing model with external platforms' delayed reporting, combining dynamic bidding with Bayesian conversion-rate estimation; published at AMLC 2025
  • Transformed ad sourcing on Pinterest from keyword-based to visual sourcing, using multi-modal AI across millions of daily pins
  • Architected production embedding and similarity-computation systems on AWS SageMaker, lifting conversion rate and advertiser return on ad spend
2021 — 2023

Data Scientist

Amazon Alexa · Seattle, WA

  • Architected an error-aware speech recognition system using Text-to-Text Transfer Transformer architectures, improving error correction rate by 41% and reducing word error rate by 17%, deployed to 100K+ daily customer interactions
  • Developed speaker-embedding models using RNNs to generate user speaker profiles for personalized speech recognition and synthetic voice generation
  • Led benchmarking of Alexa's speech recognition against industry models including OpenAI Whisper and Suno AI Bark, identifying performance gaps in agentic voice-shopping experiences
Feb – Apr 2021

Software Development Engineer

AWS CloudFormation · Seattle, WA

Developed a predictive ensemble machine learning model to estimate cloud infrastructure setup and resource-allocation time for AWS's foundational Infrastructure-as-Code service.

2017 — 2019

Data Scientist

Cisco Systems · Bengaluru, India

Built ML models to improve Cisco product quality and internal workflows:

  • Developed ensemble RNN and CNN models achieving 80% accuracy predicting software defects for Cisco routers and switches
  • Recommendation engine for identifying peer reviewers on Cisco's code review platform, using NLP
  • Unsupervised LDA topic modeling for service-request classification
  • Association mining to identify which files break on a given commit
2016

Data Semantics Intern

DataWeave Software · Bengaluru, India

Built a product-clustering algorithm across e-commerce sites for pricing insights, scaled to 10M concurrent users via distributed job scheduling. Built an SVM/random-forest/neural-net classifier that improved categorization accuracy from 81% to 90%.

2015

Big Data Analytics Intern

Zettamine Labs (Apple Inc. client) · Hyderabad, India

Built an end-to-end product that scraped and analyzed customer reviews to surface product issues for Apple — including battery-drain complaint trends. The underlying NLP research was selected for presentation at the MongoDB Conference, New York, 2015.

Skills

Machine learning

NLP, computer vision, multi-modal AI, reinforcement learning, Bayesian methods, statistical modeling, causal inference

Deep learning / AI frameworks

PyTorch, TensorFlow, Keras, LangChain, OpenAI, AWS SageMaker

Languages

Python, Java, SQL, C++, C, R, MATLAB, JavaScript, HTML, CSS

Developer tools

Git, Docker, Kubernetes, Shell, AWS, Google Cloud Platform, Jenkins, Streamlit

Education

M.S. Data Science — Indiana University Bloomington, 2019–2020
M.Tech & B.Tech Information Technology — IIIT Bangalore, 2012–2017

Publications & patent

Measurement and optimization of cross-platform advertising across Amazon Sponsored Products and Meta Ads

Amazon Machine Learning Conference (AMLC), 2025

Frameworks for real-time bidding, cost-per-click estimation, and incrementality measurement in cross-platform advertising.

System and methods for co-bidding and dynamic cost-sharing between advertisers and e-commerce platforms in cross-channel advertising

USPTO application, Amazon Patent Group — patent pending

A production system for dynamic cost-sharing in cross-channel advertising.

Second patent — details to follow

Amazon Patent Group — patent pending

NeuralCook — Image2Ingredients and cooking recommendation using deep learning

Indiana University Bloomington, 2020

Identifies ingredients from dish photos and recommends recipes, using joint NLP and computer-vision embeddings.

Intelligent defect creation using Siamese CNN-LSTM techniques

ICBAI, 2018

Duplicate bug detector for Cisco's defect tracking system, retrieving similar bugs at ~90% accuracy.

Customer success using deep learning

Advances in Economics and Business, 2018 · DOI: 10.13189/aeb.2018.060607

Deep learning models that identify customer-success patterns from behavioral data to inform business decisions.

Predicting customer-facing issues using unsupervised learning

ICBAI, 2017

Predicts post-release product issues on Cisco's next-gen devices at 95% accuracy, ahead of customer reports.

Projects

NeuralCook

Indiana University · 2020

Ingredient recognition + recipe recommendation from food photos, using joint NLP/CV embeddings. Open-source Image2Ingredients component later integrated into a Snapchat Lens in collaboration with Snap Inc.

HCI virtual receptionist

Siemens India · 2018

Three-layer virtual agent: dialogue management, video/face analysis, and speech synthesis.

Automated essay grading

IIIT-B · 2018

Custom POS tagger built with the SCRDR algorithm, feeding a neural grading model.

Carpooling web app

IIIT-B · 2017

Crowdsourced ride-sharing app for shared commutes, built in Django.