Staff Data Scientist · Seattle, WA

Mohammad R.
Khalili Shoja

Building recommendation systems and applied ML at web scale — from LLM-powered relevance models to survival-based repurchase prediction.

View my work LinkedIn
10+ Years in ML & Data Science
Ph.D. Computer Engineering · 3.93 GPA
Patent Resource-Efficient Sequence Tagging
Mohammad R. Khalili Shoja

About

I'm a Staff Data Scientist at Walmart Global Tech, where I lead homepage recommender models and architect consumables repurchase prediction systems deployed across the Walmart shopping journey.

Previously at Microsoft, I was tech lead for PA relevance — applying deep learning, NLP, and LLMs to query understanding and ads relevance at massive scale. My work spans recommender systems, ranking & retrieval, reinforcement learning for RecSys, and large-scale distributed ML on GCP and Spark.

I hold a Ph.D. in Computer Engineering from Iowa State University (GPA 3.93), with research in information theory and cryptographic key establishment — work featured in Vice.

Experience

Staff Data Scientist

Walmart Global Tech

  • Leading homepage recommender models including Cross Sell and Recommended for You, capturing user behavior signals to improve customer engagement.
  • Architecting consumables repurchase prediction models that estimate refill propensity and trigger personalized nudges across multiple ranking pipelines.
  • Designed and implemented Walmart's Discovery feature — a TikTok-style shopping experience for exploring products via short-form video content.
  • Researching GRPO and reinforcement learning to enhance long-term recommendation system performance.
  • Working with Apache Spark, GCP, and robust CI/CD pipelines at terabyte scale.

Senior Data and Applied Scientist

Microsoft

  • Tech lead for PA relevance — ML, deep learning, and NLP for query understanding and ads relevance; pioneering LLM integration for query understanding.
  • Built a comprehensive brand and product knowledge graph leveraging large language models for enhanced data accuracy.
  • Optimized small-sized LLMs (phi2) with advanced techniques to match larger models for brand graph creation at scale.
  • Led LLM adaptation for non-English languages with focus on vocabulary expansion for international markets.
  • Deployed relevance models for Edge browser price comparison and a zero-shot model for the Microsoft Store.
  • Designed a DistilBERT-based intent model using organic search results, significantly reducing customer DSATs.
  • Built end-to-end ML pipelines from data collection through A/B testing at terabyte scale.
  • Filed patent: Resource-Efficient Training of a Sequence-Tagging Model (MS 412438-US-NP).

Senior Data Scientist

Aisera

  • Designed a multi-stage unsupervised learning system for discovering and understanding novel intents.
  • Implemented novel slot filling approach using character-level information for conversational AI.

Data Scientist

Evisort

  • Built specialized Spell Checker using Seq2Seq with attention (trained on legal corpus) to improve OCR quality.
  • Constructed bi-directional LSTM-CRF with regex-based legal logic to parse contract metadata.
  • Implemented multi-class/multi-label classifier with 1D-CNN and domain-specific rules for legal documents.
  • Maintained infrastructure on GCP and AWS; built real-time model monitoring dashboard with Dash and Flask.

Research Assistant

Iowa State University

  • Introduced KERMAN protocol for secret key establishment in ad hoc networks (featured in Vice).
  • Deployed KERMAN in C with Python post-processing; introduced probabilistic graphical model (SHMM) for secret key rate analysis.

Earlier Roles

Iran · 2010 — 2014

  • Software Developer & Penetration Tester — Vice Presidency for Management: built penetration testing software on Kali Linux; deployed secure network architecture.
  • Research Assistant — Iran Telecommunication Research Center: web-based sensor alert correlation and grouping system.

Education & Research

Ph.D. Computer Engineering

Iowa State University, Ames, Iowa

Minor in Statistics · GPA 3.93/4.00 · Jun 2018

Thesis: Secret key establishment from complex correlated random processes

Selected Publications

  1. MRK Shoja, GT Amariucai, Z Wang, et al. "On the secret key capacity of sibling hidden Markov models." IEEE Trans. Information Forensics and Security, 14(2), 514–524, 2018.
  2. MRK Shoja. "Secret key establishment from common randomness represented as complex correlated random processes." Iowa State University, 2018.
  3. MRK Shoja, GT Amariucai, Z Wang, et al. "Asymptotic converse bound for secret key capacity in hidden Markov model." IEEE ISIT, 1968–1972, 2017.
  4. MRK Shoja. "KERMAN: A Key Establishment Algorithm Based on Harvesting Randomness in MANETs." Int. Conf. Security and Management (SAM'16), 2016.
  5. MRK Shoja. "Secret Common Randomness From Routing Metadata in Ad Hoc Networks." IEEE Trans. Information Forensics and Security, 11(8), 1674–1684, 2016.

Technical Expertise

ML & AI

Recommender Systems Ranking & Retrieval Reinforcement Learning Large Language Models Survival Models Propensity Modeling NLP Deep Learning

Engineering

Apache Spark Spark SQL GCP AWS CI/CD A/B Testing Feature Monitoring Distributed Systems

Tools & Languages

Python PyTorch TensorFlow Scikit-Learn DistilBERT Flask Dash

Get in touch

Open to conversations about recommender systems, applied ML at scale, and reinforcement learning for RecSys.

Email m.khalilishoja@gmail.com LinkedIn mohammad-reza-khalili-shoja GitHub khaliliShoja
Location Seattle, WA