Profile banner
Kia Khezeli

Kia Khezeli

ML researcher and AI‑safety startup founder

kia
New York City, New York
Joined May 2025
Favicon imagescholar.google.com
Favicon imagetwitter.com
Favicon imagelinkedin.com

Network

777 connections
🧠
AI/ML Researchers Academics
đź’»
Big Tech AI/ML Engineers
đź“Š
Data Scientists Analysts
🔬
Deep Tech Innovators
đź’ˇ
AI/ML Founders
đź’°
Pre-Seed VC Investors
🎯
AI/ML Product

Summary

AI researcher with a strong blend of applied and theoretical expertise in reinforcement learning, bandit optimization, and diffusion models: Kia has authored multiple peer-reviewed publications and technical contributions across these areas, and he has led teams of AI engineers to develop and deploy production-grade models and robust evaluation systems for high-stakes applications in healthcare, space, and defense. nih+1
Translates research into practical tools and startups for AI safety and reliability: In 2022 he co-founded YRIKKA to provide the trust layer for physical AI by automating testing, red-teaming, and fine-tuning visual AI models especially for long tail conditions in the real world, positioning the company to serve high-stakes domains (defense, clinical diagnostics, industrial automation). siliconangle+1
Collaborates across domains (medicine, space biology, and defense): Kia has worked with clinical teams at Mayo Clinic on causal inference platforms (CRISP), contributed to NASA-related review articles on AI and precision space health, and his startup has secured early government/DoD engagements — indicating cross-disciplinary partnerships and applied ML experience in regulated/high-risk settings. mayoclinic+2
Strong technical foundation in information theory, stochastic optimization, and sequential decision making: Educational background (B.S. Sharif, M.S. McMaster, Ph.D. Cornell) and early work on source‑channel problems, bandits, and regret bounds underpin her applied ML research and systems work. ieee+1

Work

Education

Projects

Writing

Reinforcement Learning for Clinical Applications

January 1, 2023

A review connecting reinforcement learning concepts to clinical decision‑making, discussing MDPs, reward design, and challenges/opportunities for RL in healthcare (treatment recommendation, simulation environments, safety and validation).

Favicon imagencbi.nlm.nih.gov

On information gain and regret bounds in Gaussian process bandits

January 1, 2021

Paper studying information gain and regret bounds in Gaussian process bandits — part of a line of work on theoretical guarantees for sequential decision making and bandit optimization.

Favicon imagescholar.google.com

A Source‑Channel Separation Theorem with Application to the Source Broadcast Problem (Thesis / paper)

January 1, 2014

Work on information theory topics (source‑channel separation) developed during graduate studies; representative of early technical work in communication and information theory.

Favicon imagemacsphere.mcmaster.ca