Research Engineer · ML Systems & Reinforcement Learning

Selen
Uguroglu

I build and ship ML systems — from RL and contrastive learning to multi-modal models and LLM evaluation — that run in production at the scale of hundreds of millions of users.

10+ years shipping ML at Apple, Netflix, Airbnb
Ph.D. Language Technologies, Carnegie Mellon
First-author papers at NeurIPS, ECML, AAAI · 1 US patent
RL · DPO/GRPO · contrastive learning · dialogue systems
Selen Uguroglu
Selen Uguroglu, PhD
01

About

Currently

Airbnb

Research / Machine Learning Engineer — Relevance & Search

Doctorate

Carnegie Mellon
University

Ph.D. in Language Technologies

  • Apple iOS Award for Women in Technology — granted to fewer than 10 of thousands of applicants
  • Richard King Mellon Fellowship, Carnegie Mellon
  • National Merit Scholarship
  • US Patent 20160358598 — Context-Based Endpoint Detection
  • Work featured in WIRED & TechCrunch

I am a research engineer who builds ML systems that ship — and stay in production. Over the past decade I've taken methods from reinforcement learning, self-supervised learning, and multi-modal modeling out of papers and into systems serving hundreds of millions of users.

My core technical interests span reinforcement learning and preference optimization (contextual bandits, policy optimization, DPO/GRPO, offline policy evaluation), conversational dialogue systems, self-supervised and contrastive representation learning, and active / human-in-the-loop systems for data-efficient training. I care equally about the modeling and the infrastructure that makes fast, rigorous experimentation possible.

At Airbnb I work on machine learning for search, relevance, and personalization, spanning reinforcement learning, preference optimization, and generative AI. At Netflix I pioneered the contrastive learning framework that became the foundation for several production teams, and built training infrastructure that 4×'d experimentation velocity. At Apple I shipped neural query-understanding models across every Apple device and patented a language model for speech endpoint detection.

I earned my Ph.D. in Language Technologies from Carnegie Mellon University, advised by the late Jaime Carbonell. My dissertation studied robust learning under highly-skewed category distributions, and I TA'd the graduate Machine Learning course, teaching core RL concepts — MDPs, policy gradients, value iteration, Q-learning.

I've also founded two AI companies, which means I'm comfortable owning a problem end-to-end: research, implementation, evaluation, and the messy realities of deployment. My first-author work has appeared at NeurIPS, ECML, and AAAI.

02

Experience

2024 — Present

New York, NY

Airbnb

Research / Machine Learning Engineer — Relevance & Search

  • Work on machine learning for search, relevance, and personalization, spanning reinforcement learning, preference optimization, and generative AI.
  • Build and deploy large-scale models to production, with a focus on robust evaluation and fast, rigorous experimentation.

2022 — 2024

New York, NY

Jori AI

Co-Founder / CEO

  • Built vision-language models for warehouse automation in resale and recycling — fine-tuning CLIP on custom domain-specific datasets and integrating GPT-4 for structured output generation with validation pipelines.
  • Owned the full research-to-production loop: data, modeling, evaluation, and deployment for a real customer.
  • Raised pre-seed funding and shipped the system as technology provider to one of Europe's most prominent sustainability organizations.

2018 — 2022

Los Gatos, CA

Netflix

Research Scientist — Recommendations & Search

  • Led research on contextual bandit algorithms for mobile feed ranking, launching a new personalized experience to 200M+ users — featured in WIRED and TechCrunch.
  • Pioneered a contrastive learning framework for personalization that became the foundation for 5+ production teams' models.
  • Designed and productized the algorithms behind "Because You Watched" and "More Like This," presented at a NeurIPS workshop.
  • Architected ML training infrastructure enabling a 4× increase in experimentation velocity, cutting iteration cycles from weeks to days.

2017

New York, NY

Alkymi

Co-Founder / CTO

  • Founding CTO of an enterprise AI startup in finance.
  • Researched and architected LSTM / attention-based models for structured data generation from financial text, establishing baseline performance on a proprietary benchmark dataset.

2014 — 2017

Cupertino, CA

Apple

Research Scientist — Siri & Apple Search

  • Led research on neural query understanding for Apple Search, presenting the demo to the SVP and deploying to production across all Apple devices.
  • Designed a human-in-the-loop active learning system combining uncertainty sampling with human annotation to expand Siri's intent understanding across domains.
  • Invented and patented a language model for malformed speech detection in Siri, achieving double-digit efficiency gains (US Patent 20160358598).
03

Research Specializations

I

RL & Preference Optimization

Contextual bandits, policy optimization, and preference methods like DPO and GRPO, plus offline policy evaluation.

II

Conversational Systems

Dialogue and intent understanding — from query modeling for Siri and Apple Search to LLM-driven conversational agents.

III

Active Learning

Human-in-the-loop systems and uncertainty sampling for data-efficient, continuously improving models.

IV

Self-Supervised Learning

Contrastive methods and metric learning that power personalization without exhaustive labels.

V

Transfer & Adaptation

Transfer learning and domain adaptation for robust models across shifting data distributions.

VI

Distributed Training

Large-scale distributed training and optimization to accelerate experimentation velocity.

04

Selected Work

Publications & Patents

  • Context-Based Endpoint Detection

    US Patent 20160358598 · Apple, 2015

    A novel approach to speech boundary detection using contextual language models.

  • Contrastive learning for personalization

    NeurIPS Workshop

    Self-supervised methods behind "Because You Watched" and "More Like This" at Netflix.

  • First-author research

    ECML · AAAI

    Peer-reviewed work on robust and data-efficient machine learning.

View full publication list on Google Scholar →

Technical Expertise

Methods

DPOGRPOContextual banditsContrastive learningDialogue systemsMulti-modal

Deep Learning

PyTorchTensorFlow

Languages

PythonScalaJavaSQLGo

Distributed Systems

SparkRay

Cloud & Infrastructure

GCPAWSDockerAirflowMLflow
05

Select Invited Talks

Oct 2023

Sustainability Solutions with AI

Panel Speaker

Feb 2022

AI Solution in Climate

Guest Speaker, Birol Guven's course

Jul 2021

"Your AI or Our AI"

Invited Speaker, Festival of Curiosity

Dec 2020

Similarity at Netflix

NeurIPS Expo

May 2019

Netflix PRS Conference

Invited Speaker

06

Get in Touch

"I'm always happy to talk shop with people building frontier ML systems — research, engineering, or the hard parts in between."