BIOENGINEERING
RESEARCH AND ENGINEERINGRANGEL NEUROECONOMICS LAB
As a research assistant at Caltech's Rangel Neuroeconomics Lab, I investigated whether an AI-based visual attention model (SAM) could accurately predict human eye-tracking patterns in food-choice experiments. Using data from a 2023 study, I compared observed eye movements with the model's predictions to evaluate it as an alternative to costly eye-tracking experiments.
I found no significant correlation between the two, suggesting SAM transfers poorly from the natural images it was designed for to simplified experimental stimuli.
Below are examples of the study inputs, model predictions, and my graphical analysis.
Inputs![]()
![]()
![]()
SAM Outputs![]()
![]()
![]()
Analysis
![]()
The y-axis is real eye-tracking data: how much longer participants looked at the left food than the right. The x-axis is the difference in SAM saliency between the two foods, measured within a sweeping 27-pixel window.
![]()
The y-axis tracks the difference in where real people looked first. The x-axis is the SAM saliency difference as before, here measured within a sweeping 9-pixel window.
The y-axis is real eye-tracking data: how much longer participants looked at the left food than the right. The x-axis is the difference in SAM saliency between the two foods, measured within a sweeping 27-pixel window.
The y-axis tracks the difference in where real people looked first. The x-axis is the SAM saliency difference as before, here measured within a sweeping 9-pixel window.
CAPNOGRAPHY VIA IR MEASUREMENT OF CO2
For my final project in UC Berkeley's Instrumentation in Biology and Medicine course, I worked with two classmates to design and build a prototype capnograph, a device that measures exhaled CO₂ in real time to monitor respiration.
Our design used infrared absorption to measure CO₂ concentration inside a sealed chamber (repurposed from a Naked Juice bottle). When our initial LED and photodiode circuit hit hardware limitations, we redesigned the system around an integrated dual-channel infrared sensor. By comparing its signal and reference channels, we generated a real-time capnogram.
The project pushed us to rethink our approach mid-build and combine sensor hardware with signal processing into a working biomedical device. Below are our component flowcharts, device architecture, and sample data.
Component Flow Charts
![]()
![]()
![]()
Device Structure![]()
![]()
Data
![]()
Output of the Arduino/ADC. The top line is the gas chamber signal, the middle line is the reference channel, and the bottom line is the ratio between the two.
![]()
The y-axis is the CO₂ concentration of each breath in mmHg, measured by the IR sensor. The x-axis is time in seconds. The box shows end-tidal CO₂, the peak concentration at the end of each exhale.
Output of the Arduino/ADC. The top line is the gas chamber signal, the middle line is the reference channel, and the bottom line is the ratio between the two.
The y-axis is the CO₂ concentration of each breath in mmHg, measured by the IR sensor. The x-axis is time in seconds. The box shows end-tidal CO₂, the peak concentration at the end of each exhale.
MELANIN-AWARE PULSE OXIMETRY
MAPO (Melanin-Aware Pulse Oximetry) is an independent hardware project I've been building since November 2025.
Commercial pulse oximeters systematically overestimate blood oxygen in patients with darker skin, a documented bias linked to missed hypoxemia and delayed care. MAPO is designed to correct that bias at the source, so the reading a clinician sees is accurate regardless of skin tone.
Alongside the standard red and infrared light used to estimate blood oxygen, MAPO adds a green light channel that continuously tracks melanin concentration and uses it to adjust the oxygen estimate in real time.
High- and low-fidelity flowcharts of the components, the structure of the device, and my research on MAPO's application and novelty are below.
Component Flow Charts
![]()
![]()
![]()
Device Structure![]()
Research
![]()
(Link to research)
A final project paper a classmate and I wrote for Engineering Devices for Bioengineers at UC Berkeley, proposing MAPO. The research discusses significance, defines the biosignal being measured, explores hardware options for sensing and post processing, projects clinical outcomes, compares MAPO against competing technologies, and analyzes its market viability and cost.
(Link to research)A final project paper a classmate and I wrote for Engineering Devices for Bioengineers at UC Berkeley, proposing MAPO. The research discusses significance, defines the biosignal being measured, explores hardware options for sensing and post processing, projects clinical outcomes, compares MAPO against competing technologies, and analyzes its market viability and cost.