Intelligence begins
with context.

I build systems that connect intelligent sensing, multimodal understanding, and human-centered interaction.

Why context matters
Biological vision and digital sensingAn illustrative eye and camera sensor. The marks are schematic, not a measured anatomical map.
Eye and image sensor. Schematic, not an anatomical map.

Vision

More detail.
More understanding?

A close-up reveals details. Context gives them meaning.

Human vision
Estimated visual-detail equivalent
≈576MP
LSST camera
Focal-plane pixel count
3,200MP

Human vision is not a pixel array. This wide-field estimate and camera pixel count use different measures—not a matched test of visual acuity. Sources: Roger N. Clark; Rubin Observatory.

My question: how can small visual details reveal the context of everyday interactions?

Explore MicroCam

Touch

Fine detail.
Meaningful touch?

Sensing contact is one challenge. Choosing meaningful bodily feedback is another.

Human fingertip
Grating discrimination threshold
0.94mm
GelSight sensor
Reported spatial resolution
30–100μm

Mean fingertip threshold (15 participants) and sensor spatial resolution measure different things—not a ranking of overall touch. Sources: Van Boven & Johnson, 1994; Yuan et al., 2017.

My question: how can visual and auditory events become haptic feedback that fits the situation?

Explore Ambient2Hap
A fingertip and a tactile sensorA fingertip meets a contact surface above a simplified tactile sensor grid. The drawing is conceptual, not an anatomical or engineering map.
Fingertip and tactile sensor. Schematic, not to scale.
Brain signals and possible contextsA simplified brain and one illustrative waveform connect to three memory-task context symbols: taking in, holding in mind, and recalling. These are situations to consider, not states inferred from this drawing.
Brain signals alongside taking in, holding in mind, and recalling. Illustrative waveform—not recorded data or a diagnosis.

Physiology

A signal is
not a label.

Physiological signals such as EEG can offer clues. Context helps interpret them, not turn them into certainty.

Illustrative signal
1
Possible contexts
3

1 and 3 are illustrative counts, not experimental outcomes. In a 21-participant EEG study, alpha frequency varied across memory-task phases. Source: Samuel et al., NeuroImage 2018. No diagnostic claim.

My question: how can physiological signals and contextual cues support interpretation without turning a prediction into certainty?

Explore MultiEEG-GPT Explore IntervEEG-LLM

Beyond measuring: interpret the situation, then respond with care.

A system-level approach

From observations to
meaningful interaction.

SENSE

What can we observe?

Gather relevant cues. Ask about goals and preferences.

REASON

What might it mean?

Interpret in context. Keep uncertainty visible.

ADAPT

What should change?

Offer a useful response people can accept, adjust, or decline.

People’s feedback informs the next observation.

Make room for correction, changing needs, or no action.

Explore the research

Same room. Different needs.

Scripted concept demonstration · No live sensing

Same observations, different goals. Choose a context, then compare the response.

A person, a room, and a context-aware systemAn isometric room contains a person at a desk, an environmental sensor, a camera, a local processing hub, and a lamp. Light-blue paths represent illustrative evidence, deeper blue represents interpretation, and purple marks the response path. A dashed light outline is a suggestion preview, not an applied change. Visual evidence is included. The lamp is off and unchanged.

01 / Observations

Presence is not intent.

A dim room. Someone present. The camera shows them at the desk—not what they need.

Sensing choices

Without the camera, light and presence cues remain. A stated goal needs no camera. Changed evidence requires fresh approval.

The lamp changes only with your approval.

Selected Work.

View portfolio
MicroCam research overview

MicroCam

Turns a smartphone microscope camera into a low-cost sensor for recognizing contact surfaces and their fine-grained context.

Explore the system
Ambient2Hap research overview

Ambient2Hap

Transforms visual and auditory ambient cues into somatosensory electrical stimulation, extending haptic feedback in virtual reality.

Explore the system

Research & milestones

Latest updates.

  • 2026 · ICME

    Ambient2Hap accepted to IEEE ICME 2026

    Ambient cues become electrical haptic feedback in virtual reality.

  • 2026 · Interspeech

    SpeechAgent accepted to Interspeech 2026

    An end-to-end mobile infrastructure for speech impairment assistance.

  • 2026 · Augmented Humans

    EmoDrink accepted to AHs 2026

    Physiology-informed wellbeing recommendations in mixed reality.

More updates (2)

  • 2025 · CHI

    Vision-based multimodal interface taxonomy published at CHI 2025

    A data-modality lens for designing enhanced context-aware systems.

  • 2025–2026 · Research appointment

    Research Fellow at the Smart Systems Institute, NUS

    Developed multimodal affect-sensing pipelines and LLM-supported wellbeing interactions.