Meisam Hejazinia

Experience

1. A

   Amazon · Full-time

   • Principal Applied Scientist

     Oct 2024 - Present · 11 mos·San Diego, California, United States · On-site·Current

     Driving innovation in supply chain optimization technology (SCOT) through advanced agentic LLM, machine learning, and reinforcement learning to enhance coding, reasoning, math, and writing use cases, for audit at scale, and root cause analysis.

2. F

   Full-time · 2 yrs 6 mos

   • Staff Data Scientist - Research

     Apr 2024 - Oct 2024 · 7 mos·Mountain View, California, United States
   • Lead (Staff) Machine Learning/ Artificial Intelligence SW Engineer

     May 2022 - Apr 2024 · 2 yrs·Mountain View, California, United States
3. F

   Facebook

   • Lead (Staff) Research Scientist

     Nov 2019 - May 2022 · 2 yrs 7 mos·Menlo Park, California, United States

     • Reduced the performance gap from 15% to 4% in a Multi-Party Computation privacy-preserving machine learning prototype via strategic hyperparameter tuning. • Demonstrated feasibility and potential of Federated Learning (FL) for Ad ranking in a simulated environment, achieving ~54% recovery of ad value loss, resulting in increased investment for production. • Validated the concept of user-level differential privacy in an FL solution for ad ranking by implementing and analyzing it, and demonstrating comparable performance to FL without differential privacy on multiple ad models. • Led a cross-functional team to achieve a ~2% improvement in online experiment ad value in an FL solution, extending the model's capabilities to real mobile devices. • Proved the potential for ~0.5% increase in online experiment ad value through the design and implementation of Federated Ensemble Learning and Federated Split Learning (accepted as a patent). • Crafted an external industry proposal outlining the application and system design details of Federated Learning in Ad ranking, presented to industry leaders and platforms at W3C meetings. • Spearheaded and implemented Federated Learning strategies across Statistics and Privacy, Ad Business Platform, Ranking, and AI departments. • Initiated and secured support for two early-stage research projects on Private Generative Adversarial Networks/Private transformer-based networks and Private Contextual Multi-arm Bandit, leading to a publication in ICML. • Interviewed and assessed over 40 senior and management candidates, sourced 30+ top-tier candidates, and successfully led various internal privacy-preserving machine learning summits and events, per audience feedback. Also, mentored/coached 3 interns and 6 individual contributors.

4. 2

   2 yrs 5 mos

   • Data Science Team Lead Manager TLM- Recommendation and Search Filters

     Jun 2018 - Nov 2019 · 1 yr 6 mos·Austin, Texas Metropolitan Area
   • Staff Data Scientist - Recommendation and Search Filters

     Jan 2018 - Jun 2018 · 6 mos·Austin, Texas Metropolitan Area
   • Senior Data Scientist - Revenue Management

     Jul 2017 - Jan 2018 · 7 mos·Austin, Texas Metropolitan Area
5. S

   Staples

   • Senior Data Scientist

     Nov 2016 - Jul 2017 · 9 mos·Greater Houston

     Sparx innovation lab 4th US online Retail at the time • Created a prototype experimentation platform to measure the performance of recommendation systems, utilizing a Hierarchical Bayesian Logit model in Python and a data pipeline in Apache Spark. • Decreased experiment sample size requirements by 20% through the design of an algorithm integrating the Mann-Whitney-Wilcoxon semi-parametric test, mixture log-normal model, and Monte Carlo methods, resulting in a white paper publication. • Formulated a detailed analysis and road map to enhance the measurement of promotion effects across online and offline channels. • Cut the sample size requirement for all experiments by 40% by devising an adaptive Bayesian sequential testing algorithm for continuous monitoring and control. • Designed an algorithm to power a personalized recommendation system, optimizing incremental revenue for cross-selling and up-selling by leveraging recency, frequency, monetary value, and other models.

Education

1. 
   Naveen Jindal School of Management, UT DallasDoctor of Philosophy (PhD), Quantitative Marketing and Management Science2012 - 2016
2. 
   Naveen Jindal School of Management, UT DallasMaster of Science (M.Sc.), Marketing Analytics2012 - 2015

Skills

Frequently asked

Explore more