VISIONNEURALDATA+1COMMERCE
Independent · GCC · Singapore · Malaysia · Worldwideمستقل — متاح للعمل عن بُعدSelamat datang

Engineeringintelligent systemsfor the real world.

specialty:AI / ML.computer vision.marketing automation.LCL shipping.e-commerce.AI / ML.

I engineer production-grade AI and machine learning systems for businesses that need real outcomes — computer vision at theme-park scale, intelligent e-commerce with marketing automation, and LCL freight consolidation tools that ship to enterprise clients across the GCC, Singapore, and Malaysia.

See selected workSend a brief
Scroll
0+
Years building
0M+
Photos processed
0+
Production systems
0
Industries
Expertise

What I do

Six lanes of work I ship in production. Pick one — I can talk specifics.

◆ 01

Computer Vision

Face & object detection at scale, OCR pipelines, photo-matching for theme parks, real-time inference on commodity GPUs.

◆ 02

AI / ML Engineering

Model fine-tuning, LLM integration, RAG systems, evaluation pipelines, and shipping ML to production reliably.

◆ 03

Automation & Backend

Python · Node.js · workflow automation, scrapers, internal tools, REST/GraphQL APIs, queue-based processing.

◆ 04

E-commerce & Group Buy

Group-buy platforms, multi-vendor stores, payment integration, inventory sync, and shop dashboards for SMBs.

◆ 05

Shipping & Logistics (LCL)

LCL consolidation tools, freight tracking, customs documentation, route planning, and freight cost calculators.

◆ 06

Accounting & Bookkeeping

Double-entry systems, automated reconciliation, VAT handling for GCC, monthly close, and financial reporting.

Computer VisionAI / ML EngineeringPython · Node.jsE-commerceGroup BuyLCL ShippingMarketing AutomationTheme Park PhotoSelamat datangGCC · SG · MYComputer VisionAI / ML EngineeringPython · Node.jsE-commerceGroup BuyLCL ShippingMarketing AutomationTheme Park PhotoSelamat datangGCC · SG · MY
Selected work

Selected work

Real systems, real users. Tap any project for the full case study.

FACE_MATCH99.2%~40M/SEASON
Computer Vision · Production

Theme Park Photo Engine

Match-and-deliver pipeline that processes millions of ride photos per season — face matching, watermarking, and on-demand delivery to guest portals.

PythonPyTorchOpenCVFastAPIPostgreSQL
READ CASE STUDY
5+ buyers$4820+ buyers$36$34$32★ POPULAR50+ buyers$24▲ MORE BUYERS · LOWER PRICE
Full-stack · E-commerce

Group-Buy Marketplace

Tiered group-buy mechanics — the more buyers join, the lower the unit price. Vendor onboarding, escrow, and shipping cost split across the cohort.

Node.jsReactStripeRedisMongoDB
READ CASE STUDY
CONSOLIDATED87% UTIL→ DXB
Logistics · Internal tools

LCL Freight Console

Container consolidation planner — packs LCL shipments by volume, weight, and destination port. Auto-generates customs forms and arrival ETAs.

PythonNext.jsTypeScriptPostgresRedis
READ CASE STUDY
P&L · MoM↗ +18%VAT 5%
Accounting · SaaS

Bookkeeping Suite

Double-entry ledger with bank-feed reconciliation, GCC VAT module, multi-currency support, and one-click monthly P&L for SMB owners.

Node.jsPostgreSQLReactPlaidTensorFlow.js
READ CASE STUDY
triggerscrapetransformdeliverEVENT-DRIVEN · ~600 RUNS/DAY
Workflow · Internal

Automation Pipelines

Headless scrapers, ETL flows, and event-driven internal tools that replaced manual ops work — saving teams hours every day.

PythonPlaywrightCeleryRedisPostgres
READ CASE STUDY
DEFECTYOLOV8 · 24FPS · 0.3% MISS
Computer Vision · Industry

Vision Quality Control

Defect detection on production line photos — flags anomalies, tracks per-shift quality scores, and routes flagged items for human review.

YOLOv8PythonFastAPIDockerONNX Runtime
READ CASE STUDY
VAT{"invoice_no":"INV-7842""date":"2024-11-15""vendor":"Al Falah LLC""total":4,238.50}EXTRACT · 97.4% ACCURACYAR · EN · MIXED
Computer Vision · Document AI

Document Intelligence Suite

OCR + LLM pipeline that extracts structured data from invoices, IDs, and customs forms — handles bilingual Arabic/English layouts, stamps, and noisy scans from phone cameras.

PythonDonutTesseractGPT-4VFastAPI
READ CASE STUDY
Al Aroosa Storeonlinedo you have abayas?Black AbayaAED 2453 sizes availableTap to add to cart →I'll take size MOrder #2847Paid · ships tomorrow▶ INTENTparsed▶ CATALOGmatched▶ ORDERcreatedLLM ◇VECT ◇PAY ◇WHATSAPP-NATIVE COMMERCE
Conversational Commerce · LLM

WhatsApp Commerce Engine

WhatsApp-first storefront with LLM-powered product search, cart, checkout, and order tracking — built for GCC and South Asian SMBs who already sell on WhatsApp but lack systems behind it.

PythonWhatsApp Cloud APIOpenAIPostgreSQLRedis
READ CASE STUDY
policy.mdspec.pdfwiki.mdslack.txtVECTOR SPACE3 matches"what's ourVAT policy?"Answer:VAT is 5% acrossall GCC emiratesexcept KSA whichapplies 15%.[1][2]RAG · 94% GROUNDEDCITATION-REQUIRED
AI / ML · Internal SaaS

RAG Knowledge Assistant

LLM assistant grounded in company docs — Slack/Drive/Notion ingestion, semantic search, source-cited answers, and a feedback loop that learns from corrections.

PythonOpenAIPineconeLangChainPostgres
READ CASE STUDY
Client words

What people actually said.

A small selection from Fiverr buyers across the US, UK, Canada, and Australia. Same voice, same standards, wherever the work lands.

Honestly above what I expected. Delivered clean, didn't pad the scope, didn't have to chase him once. Already booking the next gig.

[Replace · Buyer 1]
E-commerce founder · 🇺🇸 United States
Computer Vision · Photo pipeline
fiverr

Did in 4 days what my last dev couldn't do in three weeks. Just gets it. No drama.

[Replace · Buyer 2]
Operations manager · 🇬🇧 United Kingdom
Python automation · Web scraping
fiverr

Best money I've spent on Fiverr full stop. Communication is spot on and the build just works the way you'd want.

[Replace · Buyer 3]
Shopify store owner · 🇨🇦 Canada
E-commerce build · Shopify
fiverr

Mate, this guy is the real deal. Plenty of devs talk a big game — Arif actually delivers. 100% will use again.

[Replace · Buyer 4]
Agency director · 🇦🇺 Australia
AI / ML · Model integration
fiverr

Fixed a bottleneck we'd been living with for two years. In a week. Cannot recommend highly enough.

[Replace · Buyer 5]
Logistics coordinator · 🇺🇸 United States
LCL shipping · Internal tooling
fiverr

Saved us about 15 hours a week, no exaggeration. Knows GCC VAT inside out which was the dealbreaker for us.

[Replace · Buyer 6]
Finance lead · 🇬🇧 United Kingdom
Bookkeeping automation · VAT
fiverr
Experience

Where I've been

2023 — Now

CV Engineer · Theme Park Photo

Building real-time photo matching at scale — millions of guest photos sorted, watermarked, and delivered to portals every season.

2021 — 2023

ML Engineer · AI lab (inside the industry)

Mid-career stint inside an applied-AI lab — close to how production CV/ML systems actually get shipped, governed, and supported once they leave the notebook. The exposure shaped how I scope every project today: model choice is the easy part; evaluation, drift, and the human review loop are where most teams quietly lose.

2020 — 2021

Freelance · Fiverr (Top-rated)

Python automation, scraping, and small-business e-commerce builds for international clients.

Side · Ongoing

Business Admin · Shipping & Books

Hands-on experience with LCL freight ops, group-buy coordination, and bookkeeping for small importer businesses.

Get in touch

Let's build something

Cold mail welcome. I read every brief — replies within 48 hours.

hello@thearifoffice.comLinkedInGitHub
Replies within 48 hours
06/09·2024
Computer Vision · Industry

Vision Quality Control

DEFECTYOLOV8 · 24FPS · 0.3% MISS
Year
2024
Industry
Manufacturing · Quality Control
Scale
3 product lines · 24 fps real-time
Role
Computer Vision · Industry
Overview

A computer-vision quality control system that monitors production line cameras for defects in real time. Per-line fine-tuning, calibrated confidence thresholds, and an explanation overlay so operators can understand and trust the model's calls.

Challenges & Solutions
● Challenge 01

Defect variance

Each product line has different defect classes; retraining must be possible in under a day.

▲ Solution 01

Per-line fine-tuning

YOLOv8 base + lightweight per-line head; new defect class trainable in 4-6 hours from 200 samples.

● Challenge 02

False positive cost

Every false positive routes an item to manual review — backlog grows fast.

▲ Solution 02

Calibrated thresholds

Per-class confidence threshold tunable per shift; balances false positive rate to review capacity.

● Challenge 03

Line speed

Inference must keep pace with the conveyor — 24 fps minimum, no skipped frames.

▲ Solution 03

Edge inference

ONNX Runtime + TensorRT on edge GPU; sustained 24 fps on a single mid-range card per line.

● Challenge 04

Operator trust

Floor staff distrust black-box models that can't explain their calls.

▲ Solution 04

Explainable UI

Grad-CAM heatmap overlay shows what the model focused on; flag-and-review workflow lets operators correct.

By the numbers
0.0%
Miss rate
0fps
Real-time
0%
Operator acceptance
0
Product lines
Stack
YOLOv8PythonFastAPIDockerONNX RuntimeTensorRTNVIDIA JetsonBasler GigE camerasMQTTOPC-UA (PLC)GrafanaMES / ERP webhook
Continue