Hill Nguyen

AI Researcher

United States12 yrs 6 mos experience

Most Likely To SwitchAI ML Practitioner

Key Highlights

Led development of AI systems for Amazon returns.
Expert in multimodal machine learning and LLMs.
Delivered high-performance speech recognition solutions.

Stackforce AI infers this person is a Machine Learning and AI expert in E-commerce and Voice Technology.

Contact

nguyenh@amazon.com LinkedIn

Skills

Core Skills

Machine LearningDeep LearningSpeech RecognitionData Science

Other Skills

ASRAutomatic Speech RecognitionElasticSearchGenAILLMsMongoDBPredictive analysisPredictive learningStatistical analysisVisual/Language Modelalgorithm developmentcompression algorithmsconversational agentscustomer segmentationdata ingestion

About

I'm currently leading the Science team in Amazon Return & Recommerce. I and my team own the Amazon return conversational agent, which chats and nudges the customers to provide more specific return reasons. Our LLM system is an agentic system with multiple agents such as LLM orchestrator, Detail-Level Classification, Question Generator, SQL/RAG agent, Summarizer, as well as Responsible AI agent. We also optimized our conversational agent and achieved less than 1.5sec for each turn on conversation. The system already supported 300M+ returns. Please try it when returning items in Amazon.com! Other than the Amazon return conversational agent, I and my team worked on Damage Evaluation AI based on multimodal ML model/VLM/foundational model. The system relies on Visual/Language Model to ingest (1) tabular data, (2) text, and (3) image to rank and predict the most relevant damage and package type to Amazon worker during return item evaluation process. Before that, I was a Senior Applied Scientists in Amazon Alexa. I have delivered Speech Recognition for different resource-constrained hardwares, ones with runtime memory ranging from 5MB to 700MB. Examples are Alexa smart glasses, Echo Show, Echo Auto, etc.. I’m quite familiar with hardware-software co-design to train and deliver Machine Learning models running on-device for both Amazon (AZ1) and third-party (Qualcomm Snapdragon, Samsung SoC) chipset.

Experience

Amazon

2 roles

Applied Science Manager

Promoted

Jan 2021 – Present · 5 yrs 2 mos

· Establish the team’s strategic direction and roadmap for OP1 (next year) and OP3 (three-year plan).
· Manage a team of Applied Scientists, responsible for building and researching (1) GenAI and LLMs, (2) multimodal foundational model, and (3) conventional deep learning.
· Deliver conversational agents using LLMs to nudge customers to provide more specific return reasons.
o Compress the models to reduce the latency for Amazon customers.
o Research state-of-the-art LLM techniques: Retrieve-Augmented Generation (RAG), distribution shift, etc.
· Deliver a multimodal foundational model to detect Item and Package Damage from returned items using (1) images from customer, (2) images from sellers, (3) textual data from product description, and (4) tabular data from Amazon catalog.
· Deliver conventional classification ML models to detect various attributes for products sold by Amazon using BERT/xgboost/mixture of experts.
· Initialize research projects, design system architectures, and lead research to production incorporation
· Work with Product Managers to prioritize production/research tasks, and Engineering teams to bring ML models and research into production.

GenAILLMsmultimodal foundational modeldeep learningconversational agentscompression algorithms+2

Applied Scientist II

Jan 2018 – Jan 2021 · 3 yrs

Deliver Automatic Speech Recognition (ASR) models for cloud-based services such as:
o Model improvement via personalization and contextualization: leveraging personalized info, multi-turn dialog, etc.
o Large ASR with large language models (LLM) each having 2-Billion parameters
Deliver ASR models that runs on devices and neural network accelerators for different languages and devices.
o Hardware-software co-design with neural network accelerators, Develop compression algorithms for on-device ML, reducing the latency by 20% and model size by 30%. Develop on-device ML solutions that is on-par with Cloud-based technologies
o Worked with both Amazon specific hardware (AZ1) as well as third party chips (Qualcomm's Snapdragon, Samsung Soc, etc.): Alexa smart glasses (Echo Frames), Alexa for BMW/Stellatis cars (Echo Auto and Local Voice Control), etc.

Automatic Speech Recognitionneural network acceleratorshardware-software co-designcompression algorithmsSpeech Recognition

Swiss re

2 roles

Analytics Specialist II

Sep 2017 – Sep 2018 · 1 yr

Analytics Specialist

Jul 2016 – Sep 2017 · 1 yr 2 mos

Consult and shape business ML/big data ideas into feasible projects
Develop data products for internal and external clients:
o Predictive learning to improve procedure efficiency.
o Customer segmentation to improve propensity to buy and price quote.
o Trend detection to improve reaction speed and control negative effects.
o Natural catastrophe detection and forecasting to create parametric insurance product.
Analytical skill: Optimization, Statistical/Stochastic analysis + modelling, Predictive/Inferring analysis, Machine learning, Graphical modeling.
Data analytics tools:
o Proficient:
 Language: R, Python, MatLab, Mathematica.
 Tools/frameworks: Tableau, Spark, MongoDB, Stanford NLP, ElasticSearch, tensorflow/ keras, sklearn, h2o, xgboost, etc..
o Knowledgeable:
 Language: Javascript, Solidity.
 Tools/frameworks: AngularJS, D3JS, Highchart.

Predictive learningcustomer segmentationtrend detectionnatural catastrophe detectionData Science

Institute for infocomm research

2 roles

Research Scientist

Oct 2013 – Jul 2016 · 2 yrs 9 mos

2015 – 2016: Security for communication systems in smart-grid.
Main skills: Statistical/Stochastic analysis, Predictive/Inferring analysis, Matlab, R, Python.
Main contributor for problem recognition, algorithm development, and Python/R programming code.
Using statistical analysis for create highly-secure secret keys for communications.
Using predictive/inferring and statistical analysis to authenticate legitimate users using machine learning.
Data analysis based on R.
2014 – 2015: Heterogeneous networks for the fifth-generation (5G) communication system.
Main skills: Optimization, Statistical modeling, Statistical/Stochastic analysis, Machine learning, Matlab, R, Python.
Modeling 5G systems using statistical modeling.
Using reinforce learning to update operating parameters in real time.
Using statistical/stochastic analysis to predict the performance for better learning.
2013 – 2014: Green multiuser MIMO cooperative communications.
Main skills: Optimization, Multivariate analysis, MatLab.
Designing energy efficient protocols for new communication systems.
Optimizing the performance of the protocols.