AI Leader  ·  Author  ·  PhD

Dr. Selim Mimaroglu

Director of GenAI & Data Science  ·  Fannie Mae

25+ years building AI that works across financial services, energy, and enterprise software — from proof of concept to production at scale. Beyond the hype.

25+Years in AI
100+Production Models
5U.S. Patents
16+Publications
Dr. Selim Mimaroglu — Director of GenAI & Data Science, Fannie Mae, author of Enterprise AI Productivity

About

Dr. Selim Mimaroglu

Dr. Selim Mimaroglu is Director of GenAI & Data Science at Fannie Mae, where he leads AI strategy and deployment at enterprise scale across mortgage finance. He previously served as Director of Data Science & Machine Learning at Oracle, where he built the AI practice from the ground up to over 100 production models.

Over more than two decades, he has led AI initiatives across financial services, energy utilities, and enterprise software. His core conviction — and the central argument of his book — is that most AI failures are organizational, not technical.

He holds a PhD in Computer Science with an AI focus, an MIT Product Management Certificate, five U.S. patents, and more than 16 peer-reviewed publications. He is an active speaker at enterprise AI conferences and a moderator of executive discussions on scaling AI from pilot to production.

Director, GenAI & Data Science
Fannie Mae — Washington, DC Area
Former Dir., Data Science & ML
Oracle — 100+ production models
PhD, Computer Science (AI)
University of Massachusetts Boston
Product Management Certificate
Massachusetts Institute of Technology
5 U.S. Patents
Non-intrusive load monitoring & energy forecasting
16+ Peer-Reviewed Publications
AI and data science research

Book

Enterprise AI Productivity

A structured, proven roadmap for transforming your organization into a high-performing, AI-powered enterprise.

Pearson / Oracle Press  ·  Published April 2026
Enterprise AI Productivity
Success Beyond the Hype

A timely and authoritative guide bridging the gap between AI ambition and execution. Covers AI organizational design, model lifecycle management, cross-functional collaboration, and real-world deployment — with anecdotes and lessons drawn from building and scaling AI across global enterprises.

304 pages  ·  8 chapters  ·  ISBN 978-0-13-547442-6
Enterprise AI Productivity book cover — Dr. Selim Mimaroglu, Pearson/Oracle Press

Intellectual Property

U.S. Patents

Five granted U.S. patents in AI and machine learning, filed during AI practice leadership at Oracle. All focused on applying machine learning to energy intelligence at enterprise scale.

U.S. Patent 11,892,933  ·  Granted February 6, 2024
AI That Learns to Predict How Fast Data Flows Through a Network or System
Systems and methods for generating a data throughput estimation model

A machine learning system using neural networks, gradient boosting, and random forests that learns from historical patterns to estimate data throughput. Enables proactive capacity planning and performance optimization across enterprise systems.

View on Justia →
U.S. Patent 11,893,487  ·  Granted February 6, 2024
AI That Detects Whether a Specific Device Is Consuming Power at a Location — Using Only the Main Meter
Trained models for discovering target device energy usage

ML models that scan a location’s aggregate energy signal and predict whether a specific target device — such as an EV charger, HVAC unit, or industrial equipment — is present and active. Used to support energy grid demand planning without deploying per-device sensors.

View on USPTO →
U.S. Patent 11,989,178  ·  Granted May 21, 2024
Smart Meter AI That Separates a Home’s Total Power Bill Into Individual Appliance Usage
Non-intrusive load monitoring using a novel learning scheme

A novel ML model trained to disaggregate device-level energy consumption from total household usage — identifying how much electricity the fridge, washer, or water heater consumed, using only the single utility meter reading. No per-appliance hardware required.

View on Justia →
U.S. Patent 11,989,668  ·  Granted May 21, 2024
AI That Automatically Corrects Its Own Energy Estimates When Disaggregation and Detection Disagree
Non-intrusive load monitoring using machine learning and processed training data

A refinement method that cross-checks energy disaggregation predictions against supplemental device-detection signals — automatically adjusting estimates that are internally inconsistent. Enables production AI systems to self-correct without manual retraining.

View on USPTO →
Granted January 20, 2026  ·  Pub. No. 20240405548
AI That Forecasts Future Energy Consumption for Any Business or Facility Using Behavioral Segments
Time-series energy usage forecast predictions for energy consuming entities

Entity-segment ML models that learn the behavioral energy profile of each facility type or customer segment and generate accurate time-series forecasts of future energy consumption. Supports utility load forecasting, demand response planning, and grid-level resource optimization.

View on Justia →

Speaking

Conferences & Engagements

Executive moderator, panelist, and keynote speaker at enterprise AI conferences and academic venues.

2026
Moderator  ·  Executive Roundtable
“AI in Industry: Scaling Real-World Use Cases”
ODSC AI East 2026  ·  Hynes Convention Center, Boston, MA  ·  April 2026

Led a 60-minute executive roundtable with senior AI leaders on scaling AI from proof of concept to production across enterprise organizations, as part of the AI X Leadership Summit.

2026
Panelist Upcoming
“RAG Tools and Solutions: Building Effective Retrieval-Augmented Generation Systems”
Ai4 2026  ·  The Venetian, Las Vegas, NV  ·  August 4, 2026

Enterprise-focused panel on building production-grade RAG systems, covering vendor landscape, scalability, governance, and real-world use cases.

2026
Moderator  ·  Technical Roundtable Upcoming
“Beyond RAG: Next-Gen Knowledge Integration”
Ai4 2026  ·  The Venetian, Las Vegas, NV  ·  August 5, 2026

Leading discussion on the transition from standard RAG to Graph RAG, balancing context windows with dynamic retrieval, and real-time vector database indexing.

Multiple
Speaker
Oracle Customer Edge Conferences
Various locations  ·  Multiple years

Presented to enterprise customers on AI strategy, data science, and production AI deployment.

Multiple
Speaker  ·  Researcher
Academic & Research Conferences
Various academic venues  ·  Multiple years

Presented peer-reviewed research in AI and data science across academic venues. Author of 16+ published papers in machine learning, deep learning, and applied AI.

Connect

Get in Touch

Find me on LinkedIn, follow research on Google Scholar, or reach out directly.