Get in touch
Email me LinkedIn
Sabir Thapa, in black and white, looking at the camera across a conference table

hi, i'msabir.

I build the apps and signal processing behind wearable health studies at the University at Buffalo.

Research Assistant, Embedded Sensing and Computing Lab. Before that, four years of mobile apps at eSewa and Programiz.

research

three projects,
one lab.

All in the Embedded Sensing and Computing Lab at the University at Buffalo, advised by Dr. Wenyao Xu.

stride.

The StrideSense shoe: white 3D-printed upper, lattice midsole, and a cyan glow in the heel where the sensor pod sits
sensor pod.A removable pod in the heel, where the glow is. A Seeed XIAO nRF52840 with a 6-axis IMU, at 50 Hz.
printed shoe.3D-printed TPU: a one-piece upper on a lattice midsole. One pod per foot.

StrideSense · with Fitasy

walk for a minute, and each foot gets its own gait report.

The StrideSense shoe, with the sensor pod glowing cyan in the heel

sensed in the shoe.

sent over Bluetooth.

back as a gait report.

Speed, cadence, stride length, stance and swing, foot lift and symmetry, for the left and right foot.

one walk,
three screens.

The same analysis runs in a web app, an iPad app and a React Native app. A Python port of it has to match the JavaScript exactly.

simulated walkStrideSense web app gait report with speed, cadence, stride length and a gait parameter table for each foot

01 · web app

pairs both pods in the browser.

Connect, walk, and read the report. Every stride is listed with its heel strike, toe off and length.

StrideSense iPad app with a live 3D view of both shoes

02 · iPad app

shows both shoes live.

A SwiftUI app with a 3D view of each shoe, plus charts for foot lift, foot speed and stride length.

emulated podsStrideSense virtual gait lab replaying an emulated walk next to the true foot path

03 · virtual gait lab

tests changes without hardware.

Emulated pods replay a known foot path, so the algorithm can be checked before anyone puts on a shoe.

checked against motion capture

99.9%of strides found
±12msheel strike and toe off timing

Tested on about 10,000 strides from 14 people in public motion capture datasets.

Stride length lands within 2% at slow and normal walking pace.

quit.

Sense2Quit running on a phone, recording a data collection take, next to a Wear OS watch showing that smoking detection is active
on the phone.A TensorFlow Lite model spots smoking gestures in the wrist data and logs each one.
on the wrist.A Wear OS watch streams motion to the phone while detection is on.

Sense2Quit · with Columbia University and Johns Hopkins University

A smoking cessation research app, with trend charts, reminders, tips from former smokers and a couple of small games for riding out a craving.

on Sense2Quit, I built the

live demo.

watch connection check

A screen that shows the detector's window probabilities and pipeline stats as they come in. Participants now use it to check that their watch is connected.

Watch sensor stream view

data collection.

study recordings

A study mode that saves whole takes at the watch's real sample rate, and only rejects data when there's an actual gap.

Setting up a data collection session Recording a take with a countdown

updates.

from participant feedback

Google Play in-app update prompts, so every participant stays on the current build. Plus clearer login errors, a rebuilt settings menu and fixes to the tips screen.

Sense2Quit tips screen

a study with Columbia University and Johns Hopkins University.

sync.

An EmotiBit sensor board, black with two steel studs and a microSD slot, worn on a black armband
on one clock.Each board writes its samples to a microSD card against a shared clock, so students line up beat for beat.
on the arm.An optical sensor on the skin side records raw PPG, with motion from the same board.

In Sync · pilot study, in progress

do hearts fall into step when a class meditates together?

A dim classroom during a guided meditation session, students seated and wearing sensors, a meditation video on the projector screen
A pilot session: the class meditates while the sensors record.

what I built

  • A pipeline that turns raw recordings into tidy daily files.
  • 8 second windows, scored every 2 seconds.
  • A beat counts only when both PPG channels agree on it.
  • Every rejected window keeps the reason it was rejected.
  1. Two sensors

    The MAXREFDES280 and the EmotiBit record raw PPG and motion on one shared clock.

  2. One meditation

    Thirty minutes of guided meditation, compared across baseline, meditation and recovery.

  3. How they felt

    Mood, focus and NASA-TLX workload, set against how in sync each student was.

Published at

Ritik Verma, Sreyasee Das Bhattacharjee, Sabir Thapa Chhetri, Wenyao Xu. AtomSignet: Scaling Affective Reasoning with Atomic Biosignal Diffusion. BibTeX ↗

Collaborated with

  • Fitasy
  • Columbia University
  • Johns Hopkins University

All three projects are in the Embedded Sensing and Computing Lab, advised by Dr. Wenyao Xu.

In Sync is still in progress.

protocol and recordingdone
signal qualitynow
synchrony analysisnext

experience

four years of
mobile apps.
now, wearable
research.

01 · 2025 – now

Research Assistant, ESC Lab.

University at Buffalo. Apps, firmware tools and signal processing for wearable health studies, with Dr. Wenyao Xu.

02 · 2023 – 2025

Software Engineer, eSewa.

iOS for Nepal's first digital wallet, used by more than 10 million people. Swift, UIKit and Combine, Core ML models for fraud detection, and App Store releases.

03 · 2021 – 2023

Mobile Developer, Programiz.

Features, fixes and performance work on the Programiz app, part of a learn-to-code platform with over 270 million users a year.

education

  • University at Buffalo, SUNYComputer Science and Engineering · 2025 – now
  • Tribhuvan UniversityB.Sc. in Computer Science · 2019 – 2024

toolbox

  • MobileSwift, UIKit, SwiftUI, Combine, Core ML, Flutter, Kotlin, Wear OS
  • SensingBluetooth LE, IMU, PPG, nRF52840, EmotiBit, TensorFlow Lite
  • DataPython, NumPy, SciPy, pandas, matplotlib

let's talk

Have a sensor, a study or an app in mind?

Email is the fastest way to reach me.

Sabir Thapa

now

Research Assistant, ESC Lab

based in

Buffalo, New York