🎯 About

🚀 Senior Machine Learning Researcher with over 5 years of experience in delivering high-impact solutions in Machine Learning (ML), Deep Learning (DL), Reinforcement Learning (RL), and Large Language Models (LLMs). Proven expertise in architecting scalable ML models, optimizing pricing strategies, and enhancing customer segmentation. Adept at mentoring teams, collaborating on cross-functional projects, and deploying AI/ML models in production environments.

🎓 Strong academic background in probability theory, stochastic processes, combinatorics, statistical mechanics, nonparametric statistics, and Bayesian modeling. Skilled in reinforcement learning and natural language processing, with a track record of applying advanced statistical and ML methods to real-world business problems.

💰 Committed to advancing AI applications and driving organizational success with $20M+ revenue impact through innovative ML solutions.

Beyond

Senior Machine Learning Researcher

Feb 2023 – Present | Salt Lake City, UT

Click-Through Rate (CTR) Prediction for Sponsored Ads:

🚀 Built and productionized XGBoost-based CTR model replacing legacy system, improving AUC by 12% and click prediction precision by 15%.

⚡ Designed daily retraining pipeline (BigQuery + Airflow) with 50+ engineered features including product reviews, user behavior, and device type; achieved 8% CTR improvement driving $5M+ annual ad revenue.

Conversion Rate (CVR) Prediction for Sponsored Ads:

🎯 Developed Overstock's first CVR model using XGBoost to predict the likelihood of a purchase post-click; integrated with ad ranking to prioritize high-converting ads.

📊 Built labeled training pipeline joining click logs with order data; improved post-click purchase rate by 15% and advertising ROI by $3M+ annually.

AI-Powered Product Content Optimization:

🤖 Led end-to-end platform development for automated e-commerce content creation using ML and generative AI, serving 2M+ SKUs.

✅ Productionized taxonomy classification model achieving 97% accuracy, eliminating 90% of manual categorization and saving 40+ hours/week.

🔧 Built generative attribute enrichment pipeline using LLMs (OpenAI/Gemini) to fill missing product attributes, improving content completeness by 90%.

🚀 Delivered AI SKU Builder MVP enabling partner self-service; piloted with key suppliers reducing onboarding time by 80%.

⚙️ Integrated AI systems with enterprise infrastructure (Oracle DB, Product APIs) achieving near real-time catalog updates across 100K+ SKUs.

Machine Learning Coupon Personalization (MLCP):

💌 Led personalized email coupon system using hybrid XGBoost + logistic regression predicting customer purchase probabilities across discount levels.

🎯 Deployed scalable allocation algorithm optimizing discount spend for 20M+ customers daily; integrated with ESP and marketing constraints.

📈 Achieved 3.6% revenue increase and 19% profit improvement through 10M-customer A/B test, delivering $8M+ annual impact.

Dynamic ML Pricing Optimization (MLPO):

💰 Designed dynamic pricing engine for 200K+ SKUs combining LightGBM elasticity prediction with integer linear programming optimization.

🔄 Implemented two-stage price optimization (bucket selection + fine-tuning) maximizing profit under business constraints.

📊 Delivered 4-7% profit improvement with controlled 1% sales impact; boosted competitive positioning to 65% across 170K products.

Leadership & Mentoring:

👥 Participated in hiring process and onboarded 3 new data science team members, enhancing team capabilities.

🎓 Mentored intern developing Bayesian A/B testing models for skewed revenue KPIs, reducing sample size requirements by 20-40%.

Overstock.com

Machine Learning Researcher

Oct 2020 – Feb 2023 | Salt Lake City, UT

Automated Site Sale Optimization:

🧠 Developed deep neural network and transformer-based models for demand forecasting across millions of products.

⚡ Created optimization algorithms maximizing revenue/profit under budget constraints using advanced optimization and parallel processing.

📈 Improved prediction accuracy by 10% (MAPE/MAE), reduced manual processing by 80%, and increased annual revenue by $150M (6%) and profit by $10M (4%).

🚀 Deployed production systems using Docker, Jenkins, Airflow on GCP; designed and analyzed A/B tests for algorithm validation.

Additional ML Projects:

🎯 Built probabilistic SKU selection model optimizing pricing/discounts with improved transparency and 20% cost reduction.

🔍 Implemented deduplication system analyzing 3M+ SKUs using image hashing, reducing duplicates by 95% and improving efficiency by 30%.

🏷️ Enhanced competitive price matching using image/text embeddings, improving tagging precision from 30% to 70%.

Ph.D. in Mathematics, University of Utah College of Science
Aug 2016 – Jun 2020 | Greater Salt Lake City Area

• Developed mathematical models governing the evolution of systems with complex interactions in probability theory and stochastic processes.
• Built quantitative models of infinite systems to maximize energy paths, working with renowned faculty in mathematical research.
• Conducted advanced research in random walk theory and statistical mechanics with applications to real-world optimization problems.

🏆 SalesShortcut – Autonomous AI Sales Team (Grand Prize Winner)

📅 2025

🏆 Won Grand Prize at Google Cloud Agent Development Kit Hackathon (10,000+ participants); awarded $15,000, $3,000 Google Cloud credits, premium developer membership, and Google engineer mentorship.

🤖 Designed and implemented a multi-agent AI system using Google Cloud ADK and Gemini 2.0 to autonomously execute the full SDR workflow: lead discovery, research, personalized outreach, calls, follow-ups, and pipeline management.

🏗️ Architected 34 specialized agents (LLM, sequential, parallel, loop, custom) orchestrated via ADK; deployed as microservices on Google Cloud Run with inter-service A2A protocol across 5 microservices.

🔗 Integrated Google Maps, Search, Gmail, Calendar, Vertex AI, BigQuery, Pub/Sub, and ElevenLabs voice API to enable prospect targeting, tailored proposals, and real-time engagement.

⚙️ Implemented advanced orchestration patterns (including parallel processing and human oversight) enabling scalable and fault-tolerant lead processing.

🚀 Built production-ready system validated with real-world business outreach; demonstrated automation of manual SDR workflows with significant productivity gains and strong adoption potential.

🧠 Enterprise RAG Knowledge Assistant

📅 2025

🤖 Developed and deployed internal AI assistant for Beyond enabling marketing, merchandising, and supply chain teams to query pricing, promotion, product, inventory, and partner data from text documents, spreadsheets, and images.

📄 Built document ingestion pipeline (contracts, promotional flyers, pricing sheets, product images) with OCR, metadata extraction, and embedding generation.

🔍 Implemented multimodal retrieval using FAISS vector store with CLIP embeddings for images and BERT embeddings for text, ensuring high-precision responses.

⚡ Integrated with LLM backend for context-aware answers and source citation; reduced time-to-insight for business queries from hours to seconds.

📊 Delivered solution to product, marketing, merchandising, and supply chain teams, improving decision-making speed and reducing manual data lookup by 70%.

🎓 University of Utah: PhD in Mathematics, 2020
📝 Thesis: The Lyapunov Exponent of Random Walk in Random Potential
🔍 Key Contributions:

🧮 Proved a shape theorem and derived a variational formula for the limiting quenched Lyapunov exponent of random walk in a random potential on a square lattice of arbitrary dimension

⚡ Developed framework for potential functions of stationary environments with moment assumptions tied to environmental mixing

🌐 Applied results to directed and undirected polymers, and random walk in static and dynamic random environments

❄️ Extended findings to zero-temperature scenarios for last-passage percolation and first-passage percolation

🏆 GPA: 4.0

🎓 University of Utah: MSc in Statistics, 2018
📝 Thesis: Analysis of PhD applications in Mathematics
🔍 Project Overview:

📊 Analyzed comprehensive dataset of PhD applicants to major universities

🤖 Applied machine learning methods to predict admission committee decisions (accept/waitlist/reject)

🎯 Developed predictive models for rating scores assigned by admission committees

🏆 GPA: 4.0

🎓 Boğaziçi University: MSc in Mathematics, 2015
📝 Thesis: Disorder regimes of directed polymers: lattice case versus tree case
🔍 Research Focus:

🧬 Analyzed disorder regimes of directed lattice polymers versus tree polymers

⚗️ Introduced polymer chemistry models and associated random walk frameworks

🌐 Studied directed polymers in random environment with randomly distributed impurities

⚡ Modeled polymer chains affected by environmental impurities using energy correlation

🌡️ Analyzed temperature-dependent phase transitions and asymptotic behavior

🔥 Investigated high-temperature regimes where random disorder has minimal effect

❄️ Examined low-temperature phases where random disorder significantly impacts behavior

🌳 Established results for multiplicative cascades analogous to tree polymers

📊 Compared lattice and tree polymer results to address critical temperature problems

🏆 GPA: 4.0

🎓 Boğaziçi University: BSc in Mathematics, 2012
📝 Thesis: The Optimal Stopping Problem
🔍 Project Overview:

👔 Analyzed the secretary problem for optimal candidate selection under irreversible decisions

🎯 Developed strategies to maximize probability of choosing the best candidate

📊 Applied Markov chain framework with appropriate state space modeling

📚 Reviewed optimal stopping theory fundamentals and formulated Bellman equations

⚡ Solved optimization problem to derive optimal decision strategy

🏆 GPA: 4.0

KDD 2025 Publication: Debiased ML for Price Promotion Optimization 2025

2025

📚 Published in the Applied Data Science Track at KDD 2025: "A Debiased Machine Learning Framework for Optimizing Price Promotion within E-commerce".

🔬 Proposed the Delta Method, a de-meaning technique inspired by fixed-effects regression, to isolate within-product variation and estimate causal treatment effects from observational data.

📈 Demonstrated a 3% increase in revenue and 2% increase in profit via a real-world pricing experiment using the method on a furniture e-commerce platform.

• A shape theorem and a variational formula for the quenched Lyapunov exponent of random walk in a random potential - arXiv preprint (2020)

• 🏆 Intro to ML: Language Processing - AICamp (Issued Dec 2019)

🌟 Executive Director, Emerald Hills Institute
📅 Sept 2016 – Present

• 🎯 Led a nonprofit organization promoting cultural understanding and education.
• 📈 Expanded membership by 30%, engaging over 50 active members.
• 🌍 Directed initiatives benefiting over 5,000 community members.

🕊️ Board of Directors, Gandhi Alliance for Peace
📅 May 2019 – Present

• 🎯 Mission: Increase public awareness and understanding of Mahatma Gandhi, his role in world history, and his commitment to truth, nonviolence, service, and justice.
• 📚 Promote awareness of Mahatma Gandhi's values and philosophy through community outreach and educational programs.

• 📊 Generalized Linear Model project: Problem, R file, Analysis output
• 📈 Time Series Analysis project: Problem, R file, Analysis output
• 🔄 Forward and Backward Stepwise using p-values: Problem, Analysis output
• 💰 Modelling income per state project: Problem, R file, Analysis output
• 🔧 Alternative to least squares project: Problem, R file, Analysis output
• 👶 Fetal Curve Study (Quantile regression): Problem, R file,
• 📊 Beta regression and Survival Analysis: Problem, R file,
• 🎓 Special Project for Masters degree in Statistics: Analysis

• 📧 Mail: sergazy.nurbavliyev@gmail.com
• 💻 Github: github.com/sernur
• 💼 Linkedin: linkedin.com/in/sergazy/
• 🐦 X (Twitter): x.com/sergazynur
• 🧩 Leetcode: leetcode.com/u/sergazy_nur/
• 📝 Medium: medium.com/@sernur213