Julian Ramos
Julian Ramos

Julian Ramos

Senior Research Scientist · ex-Meta Reality Labs

I’m a senior research scientist and technical leader who turns open-ended questions in multimodal sensing into working algorithms, product decisions, and shipped features. At Meta, I founded and led the roll and pinch-and-drag research program, using wrist-motion and forearm-muscle signals to control Meta Ray-Ban Display glasses, personally built the core gesture software, and helped ship both interactions with the Meta Neural Band. Before Meta, I worked on applied AI research across several domains, including personalized sleep recommendation using RL and wearables (PhD thesis), hand-gesture recognition from IMU sensors, activity recognition, and stress recognition from physiological signals. My research has been published in venues like Nature, CHI, UbiComp, and PerCom.

Currently open to new roles, and particularly excited to work on embodied AI, robotics, and agentic AI.

Experience

Meta Reality LabsSenior Research Scientist · 2021–2026

Samsung Research AmericaResearch Intern · 2016

Carnegie Mellon UniversityRobotics InstituteMachine Learning Engineer · 2008–2010

Carnegie Mellon UniversityHuman–Computer Interaction InstituteMachine Learning Engineer · 2010–2015

Education

Carnegie Mellon UniversityPh.D., Computer Science · 2015–2021

Carnegie Mellon UniversityM.S., Human–Computer Interaction · 2015–2019

Universidad Militar Nueva GranadaB.Eng., Mechatronics Engineering · 2000–2006

Shipped work

Meta Reality Labs

Interactions shipped with Meta Ray-Ban Display and Meta Neural Band.

Thumb-swipe scrolling · 1:47 ↗

Thumb swipes: subtle navigation

Contributed extensively to thumb-swipe interaction through gesture research, data-collection tools, onboarding, and human evaluation—helping translate neural input into usable navigation.

Pinch-and-drag · 3:30 ↗

Pinch-and-drag: reliable navigation

Led research and product improvements to gesture reliability, visual feedback, and control-display gain, making navigation more efficient.

Research leadership

From evidence to decisions

At Meta, I turned undefined opportunities into research programs and shipped features. I established research directions, defined hypotheses and success criteria, and set roadmaps. As one example, I founded and led the IMU gesture research program, including roll and pinch-and-drag, and built its core research stack, giving teams a shared foundation for rapid prototyping and evaluation. I used evidence from user studies to guide investment, refine interactions, and move the work into product. The selected examples below highlight my most notable contributions.

Closed-loop gesture assessment

Neural-band gesture-set exploration and validation

I designed, implemented, and evaluated candidate gesture sets, then used participant performance to select the set that shipped. The video shows an early tool we built to evaluate EMG model performance in real time, with participants using gestures to play a game. This work is part of a Nature paper.

3 studies · ~750 participants

Co-adaptive motor-unit control

Prototype-driven evaluation

I turned open interaction questions into working prototypes, research pipelines, and human evaluations. This made it possible to test unconventional ideas early and decide which directions warranted more work.

11+ prototypes · 4 studies · 178 participants

Wrist-rotation targeting evaluation

Reusable evaluation tools

I built data-collection software and reusable research infrastructure for wearable ML and interaction studies. The video shows a tool used to measure user performance when targeting through wrist rotation.

Acquisition error by gainFive boxplots show acquisition error distributions for gain values from 1.5 to 3.5. The medians rise through gain 2.0 and then flatten into a plateau.0123456789101.52.02.53.03.5gainacquisition error
Acquisition error by gain

From study to product

I carried study findings through iteration, improving gesture reliability, visual feedback, and control-display gain until the changes were ready for product.

4 studies · 210 participants

Selected research

All publications ↓
Wrist sEMG sensing hardware, gesture decoding, and computer-input tasks from the Nature paper.
Nature · 2025

Neuromotor interfaces

Co-author of a non-invasive neuromotor interface that generalizes across people for computer input.

Read more ↗
Personalized mobile-health framework connecting user feedback, wearable sensing, and adaptive recommendations.
Doctoral thesis · 2021

SleepU

A reinforcement-learning system that sensed context, decided when and what to recommend, and adapted from human feedback to personalize sleep recommendations.

Read more ↗
Five subtle finger gestures performed while wearing an off-the-shelf smartwatch.
CHI · 2016

Serendipity

First IMU + machine-learning hand-gesture recognition on a standard commercial smartwatch.

Read more ↗

Projects

Human-Computer Interaction InstituteRobotics InstituteCarnegie Mellon University

The following is a summary of my PhD and pre-PhD work at CMU.

Physiological sensing equipment and heart-rate measurements
Physiological sensing · Machine learning

Stress Recognition

This work introduced a machine learning model for detecting stress episodes from physiological response changes under unknown physical activity.

Explore project →
Workflow connecting data samples, exploration, and user insights
User behavior · Data exploration

Other projects

Understanding smartphone behavior and designing better ways for people to explore large datasets.

Explore project →

Publications

2025

2021

2020

2019

2018

2017

2016

2015

2014

2012

2011

2010

2007