Ezitech Case Studies: Healthcare, Education, ERP and POS

EZI·B19·MERN·014

Areeba Naveed

MERN Stack Internship

Batch: B-19 · 2025
Mentor: Hamza Tariq
Client: Ezitech (in-house)

ReactNodeExpressMongoDBJWT

W0112-week trackW12

ScopedFirst PRLive

Attendance that closes its own books

Problem

EZILMS tracked intern attendance in a shared spreadsheet. Across 400+ interns and 34 tracks, mentors spent roughly six hours a week reconciling it, and stipend eligibility was only settled after the cohort had already ended.

Approach

Areeba built an attendance service on top of the existing LMS: QR check-in per session, a mentor dashboard scoped by role, and an eligibility rule engine that flags anyone below the 80% threshold at week 8 — while there is still time to fix it.

Weekly reconciliation: 6h→25 min
Attendance disputes: ↓71%
Cohorts running on it: 4

Shipped EZILMS — Attendance & Eligibility module

EZI·B18·FLTR·007

Bilal Ahmed Khan

Flutter Development Internship

Batch: B-18 · 2025
Mentor: Sana Iqbal
Client: Ezi Taxi

FlutterDartFirebaseHiveMaps SDK

W0112-week trackW12

ScopedFirst PRLive

Keeping the driver on the map at 2G

Problem

The Ezi Taxi driver app dropped trip location updates in Rawalpindi's low-coverage sectors. Dispatch would lose a driver mid-trip and the fare had to be settled by phone.

Approach

Bilal rewrote the location layer around an offline queue — points buffer locally in Hive and flush in batches on reconnect — and moved tracking into a foreground service so it survives Android's background execution limits.

Trips tracked end to end: 82%→97.4%
Cold start: 4.1s→1.9s
Battery per 8h shift: ↓34%

Shipped Ezi Taxi Driver — v3.2, Play Store

EZI·B17·ML·022

Hira Shahzad

Machine Learning Internship

Batch: B-17 · 2024
Mentor: Dr. Usman Raza
Client: Ezitech Sales

Pythonscikit-learnpandasFastAPI

W0116-week trackW16

ScopedModel v1Live

Ranking 1,100 leads a month for a three-person team

Problem

Every inbound enquiry entered the same undifferentiated queue. With roughly 1,100 leads a month against three people, the order they were worked in was effectively random — and the best leads cooled while someone answered a tyre-kicker.

Approach

Hira labelled fourteen months of CRM history, trained a gradient-boosted scorer on nineteen features (source, service requested, response lag, organisation domain, page depth), served it behind FastAPI and wrote the score directly into the CRM list view so nobody had to open a second tool.

Contact → qualified rate: 11%→19%
First response, top decile: 9h→40 min
Held-out AUC: 0.83

Shipped Lead scoring service — running in production since Nov 2024

EZI·B19·UX·003

Maryam Fatima

UI/UX Design Internship

Batch: B-19 · 2025
Mentor: Zainab Ali
Client: Auto parts retailer

FigmaMazeDesign tokens

W0110-week trackW10

ResearchPrototypeHandoff

Eleven taps to four, at the counter

Problem

EZIPOS checkout took a cashier eleven taps to close a routine three-item sale. New counter staff needed about two weeks before they stopped slowing the queue.

Approach

Maryam ran nine counter-side observation sessions during real trading hours, then rebuilt the flow around a persistent numeric pad and a single-screen cart. She also shipped a 40-token design system so the dev team stopped re-deriving spacing on every screen.

Taps per standard sale: 11→4
Average sale time: 48s→22s
New-cashier ramp: 2 wks→3 days

Shipped EZIPOS checkout redesign + Ezi design token set

EZI·B18·LARA·031

Usman Javed

PHP & Laravel Internship

Batch: B-18 · 2025
Mentor: Faisal Mehmood
Client: Private hospital, Rawalpindi

Laravel 11MySQLLivewireRedis

W0112-week trackW12

ProfiledRefactorLive

The month-end invoice run that used to time out

Problem

The EZIERP billing screen at a client hospital failed every month-end. The invoice query scanned 1.4 million unindexed rows, and accounts closed the month by exporting to Excel instead.

Approach

Usman profiled the request path with Telescope, added the composite indexes the query actually needed, introduced a materialised daily-ledger table, moved PDF generation onto a Redis-backed queue and paginated the ledger view.

Month-end invoice run: 4m 20s→6s
Timeouts at close: every month→0
Peak server memory: ↓61%

Shipped EZIERP billing module — deployed to 2 hospital sites

EZI·B17·SEO·009

Zoya Rehman

SEO Internship

Batch: B-17 · 2024
Mentor: Ahsan Sheikh
Client: US medical billing firm

Search ConsoleAhrefsScreaming FrogSchema.org

W0112-week trackW12

AuditBuildIndexed

Ranking for the service, not just the brand name

Problem

The client ranked for its own name and almost nothing else — 210 organic sessions a month in a market where competitors were publishing against 2,400 service-intent pages.

Approach

Zoya mapped 84 service-intent keywords onto twelve new landing pages, cleared 340 crawl errors and a duplicated title template that was collapsing the service tree, and added FAQ and LocalBusiness schema across it.

Organic sessions / month: 210→1,860
Keywords in top 10: 3→47
Featured snippets held: 2

Shipped 12 service landing pages · technical fixes · schema layer

EZI·B19·CV·011

Ali Hassan

Computer Vision Internship

Batch: B-19 · 2025
Mentor: Dr. Usman Raza
Client: Driving school, Islamabad

YOLOv8OpenCVONNX RuntimeJetson Nano

W0116-week trackW16

DatasetTrainedOn site

Reading number plates at the test-track gate

Problem

Vehicle entries at the test track were written into a paper register. On roughly one day in five the register did not reconcile with the booking system, and disputed slots had to be resolved from memory.

Approach

Ali collected and annotated 3,200 plates photographed on site — covering both Urdu and Latin plate formats and the angled mounting of the existing gate camera — fine-tuned YOLOv8n, added a perspective-correction stage, then exported to ONNX and ran it on a Jetson Nano at the gate.

Plate read accuracy: 94.1%
Gate log reconciliation: manual→automatic
Shifts freed from data entry: 2

Shipped Gate ANPR unit — running on site since Jun 2025

EZI·B18·GAME·005

Talha Siddiqui

Game Development Internship

Batch: B-18 · 2025
Mentor: Kashif Nadeem
Client: Ezi Ludo (in-house)

UnityC#Addressables

W0114-week trackW14

ProfiledRebuiltReleased

Getting Ezi Ludo onto a 2 GB phone

Problem

The build weighed 148 MB and fell to 24 fps during the four-player board animation — on exactly the entry-level Android devices that make up most of the install base.

Approach

Talha moved boards and audio to Addressables with on-demand loading, replaced per-frame instantiation with an object pool for tokens and dice, and baked the board lighting instead of computing it live.

Install size: 148 MB→61 MB
Sustained frame rate, 2 GB device: 24 fps→58 fps
Day-1 crash-free sessions: 91%→99.2%

Shipped Ezi Ludo — v2.0, Play Store

EZI·B19·BRND·018

Nimra Aslam

Brand Identity Design Internship

Batch: B-19 · 2025
Mentor: Zainab Ali
Client: Bakery & dessert brand

IllustratorFigmaPantone

W0110-week trackW10

AuditMarksGuide

One mark, six places it had to survive

Problem

Three different versions of the logo were live at once — signage, packaging and Instagram each carried a different one. Print vendors had started re-drawing the mark by eye, and no two batches matched.

Approach

Nimra rebuilt the mark as clean vectors with a four-size responsive lockup set, fixed the palette to CMYK with Pantone equivalents for print, and delivered a 24-page identity guide alongside a print-ready asset pack.

Approved marks in circulation: 3→1
Vendor re-draw requests: →0
Later campaigns reusing the pack: 3

Shipped Identity system · 24-page guide · print asset pack

EZI·B18·SMM·026

Hamza Rauf

Social Media Marketing Internship

Batch: B-18 · 2025
Mentor: Ahsan Sheikh
Client: Ezitech Admissions

Meta AdsMetricoolGA4Canva

W0110-week trackW10

BaselineCreativeScaled

Filling 400 seats without relying on referrals

Problem

Internship intake ran almost entirely on word of mouth — around 90 applications per cohort against 400 seats — and the paid spend that did exist had no attribution behind it.

Approach

Hamza collapsed 34 tracks into eight creative clusters so each ad spoke to a real audience, wired UTM and lead-form attribution end to end, then rebalanced budget weekly against cost per qualified application rather than clicks.

Applications per cohort: 90→610
Cost per qualified application: Rs 412→Rs 137
Tracks filled from paid alone: 3

Shipped Admissions acquisition system — live for 3 cohorts

Client projects we built

Attendance that closes its own books

Problem

Approach

Keeping the driver on the map at 2G

Problem

Approach

Ranking 1,100 leads a month for a three-person team

Problem

Approach

Eleven taps to four, at the counter

Problem

Approach

The month-end invoice run that used to time out

Problem

Approach

Ranking for the service, not just the brand name

Problem

Approach

Reading number plates at the test-track gate

Problem

Approach

Getting Ezi Ludo onto a 2 GB phone

Problem

Approach

One mark, six places it had to survive

Problem

Approach

Filling 400 seats without relying on referrals

Problem

Approach

Your work could be the next record on this page.

More of our work

Address:

Office #304-G, Amna Plaza, Main Peshawar Rd, Rawalpindi, Punjab

Contact:

Phone and WhatsApp 24/7: +92 345 5555396

info@ezitech.org

Company

Services

Products

Industries

Resources