01 Customer
EMR auto-refresh
A Chrome extension that reloads the EMR daysheet the moment a patient's status changes, so the front desk is never acting on a stale board.
available · ontario, canada
Shahmeer Shahid
Forward Deployed Engineer
I do my best work close to the problem: sitting with the people at a crossroads, mapping how a fix should work, then writing the code and putting it in their hands. Lately it's all healthcare, from clinic-floor automation to quantum ML research.
some
selected
in page
selected
PROJECTS
livein page
Customer work, personal projects, some research and a playground.
A day at the clinic: real floor plan, live boards, the door-to-door time auto-logged as data. Fictional patients.
Manual Entry vs. Automated Watcher
One biometer printout · eight EMR fields buried in the noise · two ways to move them.
Manual Entry
Staffer reads the paper printout
Time 00:00.00
Keystrokes 0
Errors 0
Biometer
NAME Carl Shen:)
MAY 25 2026 9:28AM
NO:51112 SN:AW7000633
REF.DATACYL:(−)
VD:13.75 PD:59
<R>SCA
−1.25−0.2585
−1.00−0.2583
−1.25−0.2587
▸−1.00−0.2585
S.E.−1.25
<L>−0.50−0.5082
−0.50−0.5084
KRT.DATA
<R>DMMA
R144.007.67101
R245.507.4311
AVG44.757.55
R1−R2:−1.50
<L>44.007.6871
45.007.50161
TONO.DATAmmHg AVG
R141516
L(15)1616
PACH.DATAµm AVG
R495498497
L505500502
TOPCON
EMR · OD Encounter 0 / 8 filled
Sphere (OD)
D
Cylinder (OD)
D
Axis (OD)
°
PD
mm
K1 (OD)
D
K2 (OD)
D
IOP (OD)
mmHg
CCT (OD)
µm
Watcher
Script reads the digital export
Time 00:00.00
Actions 0
Errors 0
Biometer
EMR · OD Encounter 0 / 8 filled
Sphere (OD)
D
Cylinder (OD)
D
Axis (OD)
°
PD
mm
K1 (OD)
D
K2 (OD)
D
IOP (OD)
mmHg
CCT (OD)
µm
Watcher advantage 0×faster
Time 00:00 vs 0.42s
Keystrokes 0 vs 1
Errors 0 vs 0
workflow benchmark · 01
One biometer scan, two ways into the chart.
A Topcon Omnia captures a patient's biometry as a dense printout, but those eight values still have to land in the EMR. Here are two ways to move them, racing the same scan.
Manual entry. A staffer reads the paper printout and types all eight fields by hand, typos and all.
omnia-to-oscar watcher. A script reads the machine's digital export and writes the chart in one pass.
Classical ML vs. Quantum ML
Same MedMNIST chest X-rays. Two models. One honest benchmark.
Ready
encode → train → evaluate Epoch
00 / 20
Classical acc
50.0%
Quantum acc
50.0%
MedMNIST · PneumoniaMNIST 0 / 36
Hover any sample to inspect what each model predicted.
Test accuracy vs. epoch
Classical CNN · 121,410 p Hybrid QNN · 4 qubits · 96 p
Trainable parameters · √-scaled
CNN 0
QNN 0
No breakthrough yet. The quantum model reached 91% against the classical model's 96%, using roughly 1,300× fewer trainable parameters, yet it trained far slower and stays capped by today's qubit counts. The opening, if there is one, is parameter efficiency on small medical datasets, not raw accuracy.
Research with Dr. Edward Sykes · Quantum Computing Lab, University of Guelph qclab.uoguelph.ca ↗
research · breadth
Quantum vs classical, on real medical images.
With Dr. Edward Sykes at the Quantum Computing Lab in Guelph: two models train on the same MedMNIST chest X-rays, then face an honest verdict. The interesting result is not who wins on accuracy.
Classical CNN. A conventional network with about 121,000 trainable parameters, the accuracy to beat.
Hybrid QNN. A four-qubit quantum model with 96 parameters, chasing the same accuracy on a fraction of the budget.
more builds
More of what I've built.
Most quietly take a load off a real clinic workflow, niche on purpose and relied on every day. The last one is just for the fun of it. All of them got shipped and have real users.
02 Customer
Clinic status boards
Live boards the desk and doctors keep open all day, read straight from OSCAR: a doctors' lounge queue, an OCT triage board, the vision and HVF rooms. Cards move themselves as patients check in and progress, so everyone sees the same picture.
03 Customer
The device share
A single network share that consolidates exports from roughly a dozen diagnostic machines into one place. The foundation the downstream automations depend on.
04 Customer
Clinic hub
One internal landing page that pulls the clinic's tools into a single starting point, the boards, the order forms, the shared exports, so staff are never hunting for a link.
05 Customer
Clinic server
An always-on server for the clinic that launches the tools each morning, keeps them updated from git, and powers down at night to save resources.
06 Customer
Tri-City Eye Care site
A referral-first fresh build of the clinic's website.
07 Playground
eyecase
A daily ophthalmology case puzzle: one case, one guess, a leaderboard. A focused, fun teaching tool that runs entirely in the browser.
the path
Solving floor-level problems from the start.
Bombardier
A foreign-object-debris detector built on Raspberry Pi and OpenCV, spotting tools and goggles left inside an aircraft, plus automated weekly reporting on non-conformance trends.
Onward Manufacturing
A product-trained assistant for Broil King and a guided, multi-step FAQ that walks customers to the right grill, front to back in React.
Mitsubishi Heavy Industries
Shipped a company-wide intranet (Next.js, Payload, MongoDB) where staff reach their resources, internal communication, and onboarding in one place. Earlier, time on the production floor riveting ribs led to building a computer-vision model that flags a faulty rib.
Tri-City Eye Care
Bringing what I've learned into a clinic, where the software I ship has a visible impact, turning manual work into tools the staff actually rely on.