Research-Focused ML Engineering Leader
Alok Upadhyay Building intelligence at scale.
Hands-on, research-focused ML Engineering Leader with 12+ years at Amazon & AWS. Leading 20+ person orgs building Multimodal AI products, large-scale Recommendation Systems, and Generative AI pipelines serving millions of users. 4 US Patents. Published at ICLR. IEEE Senior Member.
12+
Years at Amazon
20+
Current Team Size
3B+
Messages / Month
1B+
Person Recognitions / Day
4 US Patents Granted
Identity, Authentication & Multi-Modal Recognition
Writing
Blog
Thoughts on building Multimodal AI, ML systems at scale, and lessons from the trenches.
Loading posts...
Career
ML Engineering Leadership Experience
From founding engineer to Manager of Managers -- building hands-on, research-driven ML products at massive scale.
Software Development Manager
Amazon -- Prime Video Personalization & Discovery
Jul 2024 -- Present
- Lead the Outbound Message Generation & Engagement organization (~20 engineers, SDMs, and TPMs), owning customer enrollment, engagement, and retention through personalized touchpoints
- Architected a self-hosted GenAI Auto-QA system on AWS Bedrock using LLMs to validate 500M+ assets/month across Email, Push, and WhatsApp, eliminating manual vendor review
- Direct a massive-scale content generation pipeline delivering 3B+ personalized messages monthly, leveraging transformer recommendation architectures for retrieval and ranking
- Established the technical roadmap transitioning the org from legacy systems to a GenAI-first architecture, reducing inference cost by 17% through model optimizations
Software Development Manager
Amazon AGI -- Ambient Recognition & Authentication
Mar 2020 -- Jul 2024
- Led a cross-functional team of ~14-18 SDEs, SDMs, Applied Scientists, and TPMs; managed the Manager of Managers layer for Authentication/Authorization/Presence teams
- Launched the Multimodal Recognition Engine fusing VoiceID, VisualID, PhoneID, and usage patterns to infer user identity, with cloud-edge synchronization in under 8ms
- Delivered VisualID (Facial Recognition) and PhoneID (Bluetooth Proximity) for Alexa personalization, coordinating across 23+ external teams
- Built the ML training pipeline for Continuous Improving Multimodal Recognition -- an online learning system integrated with LLM-powered Alexa+ experiences
- Devised the Authentication Confidence Levels (ACL) security standard -- a 6-tier scoring scheme adopted company-wide
Senior Software Engineer
Amazon AGI -- Ambient Recognition & Authentication
Nov 2018 -- Mar 2020
- Founding Engineer; architected high-stakes identity and security features resulting in multiple granted US Patents
- Architected "Limit Access" on Alexa -- a multi-factor authentication system (Voice PIN + VoiceID) achieving HIPAA compliance for Amazon Pharmacy
- Designed the Person Recognition Identity API for 3rd-party developers, enabling secure personalization of Alexa interactions
- Invented the Cross-Modal Automated Ground Truth scheme (patented) using BLE proximity to ground-truth voice prints, bringing in 100M+ monthly labels
Software Development Engineer
Amazon -- Alexa Identity
May 2017 -- Nov 2018
- Founding engineer for Alexa Identity; built foundational identity services, human-in-the-loop ground truthing, offline/online ML pipelines, launched Alexa Voice Training & Recognition
- Established operational excellence mechanisms ensuring low-latency, high-availability voice biometric services at scale
Software Development Engineer
Amazon Web Services -- IAM
Aug 2016 -- May 2017
- Built the Resource Groups Tagging API enabling tag-based access control (TBAC) at 40K+ TPS
- Redesigned Tagging Discovery Services eliminating single-point-of-failure; deployed to AWS GovCloud with air-gapped security protocols
Software Development Engineer
Amazon -- Home Services
Mar 2014 -- Aug 2016
- Founding engineer for Amazon Home Services marketplace; designed seller onboarding flows, ASIN search, and AWS Step Functions workflows
- Built a hyperlocal seller notification mechanism reducing customer order claim time by 95% (~12hrs to ~30min)
Media
Featured In
Alok's work on Alexa identity, voice recognition, and ambient AI has been covered by 10+ major outlets.
Wall Street Journal
Good Morning America
TechCrunch
The Verge
CNET
TIME
ZDNet
Live Mint
Business Insider
Amazon Science
AWS Blog
Wall Street Journal
2017
Alexa Voice Recognition Launch Coverage
Alexa learns to distinguish between different voices -- the first voice biometric personalization system on a consumer device at scale.
Watch →
Good Morning America
2019
Alexa Voice PIN & Authentication
Multi-factor biometric authentication on Alexa enabling secure transactions for healthcare and purchases.
Watch →
TechCrunch
2017
Amazon Alexa Devices Can Finally Distinguish Between Different Voices
Coverage of the voice recognition launch that drove 1.1M explicit speaker enrollments in the first 3 months.
Read →
TIME
2017
Amazon Echo Voice ID Feature
TIME's coverage of Amazon Echo learning to identify individual users by voice.
Read →
The Verge
2017
Amazon Echo Multi-User Voice Recognition
The Verge on Alexa's new ability to recognize multiple users on a single device.
Read →
CNET
2021
On Echo Show 15, Alexa Will Recognize Your Face Thanks to Visual ID
Launch of Alexa Visual ID -- facial recognition for personalized experiences on Echo Show devices.
Read →
ZDNet
2019
Alexa Skills Can Now Recognise Your Voice to Personalise Services
Skills Personalization launch enabling 100+ third-party skills to deliver voice-personalized experiences.
Read →
Amazon Science
2021
Ambient Intelligence Will Accelerate Advancements in General AI
Amazon Science article covering multi-modal person recognition and ambient intelligence -- powered by the Presence Data patent.
Read →
Business Insider
2015
Amazon Introduces Amazon Home Services
Launch of Amazon's first hyperlocal marketplace, available in 2.4 million zip codes at launch.
Read →
Expertise
Multimodal AI, RecSys & GenAI Skills
Leadership & Strategy
Manager of Managers
Org Design
3-Year Roadmaps
Science-to-Production
Hire & Develop the Best
Cross-Org Programs
Zero-to-One Products
Career Development
Machine Learning & AI
Multimodal ML
Recommendation Systems
LLMs & Fine-tuning
GenAI Systems
RAG Pipelines
RLHF
PyTorch
SageMaker
HuggingFace
LangChain
FAISS
Engineering & Infrastructure
Distributed Systems
40K+ TPS
Cloud-Edge (<8ms)
AWS
DynamoDB
Bedrock
CI/CD for ML
HIPAA Compliance
Java
Python
C/C++
Intellectual Property
US Patents in AI & Identity
4 granted US patents and 1 pending -- spanning multimodal identity recognition, biometric authentication, and person detection.
USPTO Pending
Multi-Modal Person Recognition
Pending
Filed Mar 2023
Research
Peer-Reviewed ML Research Publications
Published research at ICLR, MathAI, and IEEE on multimodal AI reasoning, LLM recommendation systems, and biometric geometry.
ORCID: 0009-0003-6892-48752026
Are VLM Identity Judgments Logically Consistent? Evaluating Symmetry, CoT, and Transitivity in Person Re-ID
ICLR 2026 Workshop on Logical Reasoning of LLMs
Read Paper →
2026
Do LLM Recommenders Obey Preference Axioms? Testing Logical Rationality in LLM-Based Recommendation
ICLR 2026 Workshop on Logical Reasoning of LLMs
Read Paper →
2012
A Novel Architecture for Secure Communications in Mobile Systems
Int'l Conf. Internet Technology & Secured Transactions
Read Paper →
2026
Riemannian Geometry of Multimodal Biometric Embedding Spaces
Conference of Mathematics of AI (MathAI 2026) -- Oral
Read Paper →
Soon
Additional papers under peer review
TMLR, ACM ICMR 2026, CVPR 2026 GenBio Workshop, ECCV 2026
Service
Conference Reviewer & Program Committee
Program Committee member and peer reviewer for top-tier ML and AI conferences including ICLR, CVPR, and ACM ICMR.
ACM ICMR 2026
International Conference on Multimedia Retrieval
4 papers reviewed
ICLR 2026
Logical Reasoning of LLMs
3 papers reviewed
ICLR 2026
AI in the Wild
7 papers reviewed
ICLR 2026
Multimodal Intelligence
5 papers reviewed
CVPR 2026
Foundational & Generative Models in Biometrics
Reviewer
Education
Academic Background
Birla Institute of Technology and Science (BITS), Pilani
Master of Science (Tech.), Information Systems
2009 -- 2013
Beyond Work
Certifications & Memberships
FAA Certificated Private Pilot
Single Engine Land (ASEL)
Since Jan 2022Professional Associations
Senior Member
IEEE
Member
British Computer Society (BCS)
Member
Airplane Owners & Pilots Association (AOPA)
Connect