01 / AI & ML
AI & Machine Learning Systems
RAG pipelines, vector search with CLIP embeddings, document understanding, computer vision, LLM orchestration. Production deployments, not notebooks.
PythonOpenAIGeminipgvectorCLIP
Product studio · AI-accelerated delivery · Cape Town → the Gulf
Software systems that ship, not slide decks.
One senior engineer, AI-accelerated: the output of a small team, shipped by the person you brief. W-Enterprises designs and delivers production AI/ML and full-stack systems from specification to production in around three weeks, including bilingual Arabic and English platforms for the Gulf. Out of that practice now come four productized tools: trademark clearance, document extraction, document automation and a private knowledge layer.
01 / Practice
What we actually build.
02 / Engineering
Full-Stack Engineering
Type-safe applications on Next.js, React, Vue, Node, .NET and Go. Serverless-first deployments on Vercel, Cloudflare, Firebase. Speed without sacrificing correctness.
Next.jsReactTypeScriptGo.NET
03 / SaaS
SaaS Platform Build-Out
Multi-tenant architectures, subscription billing, identity, admin tooling, regional payment integrations. End-to-end from specification to production within compressed delivery windows.
FirebaseFirestoreGeideaMoyasarStripePostgreSQLMulti-tenant
04 / Discovery
Technical Discovery & Documentation
Solution architecture, PRDs, technical specifications, requirements traceability. The formal paperwork that lets serious engagements move forward with confidence on both sides.
ArchitecturePRDsSpecificationsMermaid
02 / Built for the Gulf
Built for the Gulf, not just translated for it.
Bilingual Arabic and English platforms with correct right-to-left interfaces, shipped for the Saudi and Gulf market. PDPL data protection and regional payments (Geidea and Moyasar) are handled as first-class requirements, and invoicing flows are built with Saudi ZATCA e-invoicing requirements in mind.
هندسة برمجيات وذكاء اصطناعي للسوق الخليجي
Bilingual
Arabic / English + RTL
Native AR/EN interfaces with correct right-to-left layout, bidirectional text and Arabic typography. Not a translate widget.
Compliance
PDPL-ready
Data-protection-aware architectures: sensitive fields handled to Saudi PDPL expectations by design.
E-invoicing
ZATCA-aware
Invoicing and document flows built with Saudi ZATCA e-invoicing requirements in mind.
Payments
Geidea, Moyasar + regional
Saudi payment integration with Geidea (live hosted checkout, including Apple Pay) and Moyasar, alongside Stripe and PayPal for cross-border billing.
Full Saudi and Gulf capability → ZATCA Phase 2 integration, Wave 25 deadline 1 Feb 2027 → How correct Arabic PDFs are actually built → AI document extraction → اقرأ بالعربية ←
- / Principles
How the practice operates.
i.
One senior engineer.
No account managers. No project handoffs. The person you brief is the person who writes the code, designs the architecture, and ships the system.
ii.
Production-first.
Every deliverable lands in production with monitoring, observability, security review and CI in place, not a demo on a developer's laptop.
iii.
Documented.
Every engagement comes with formal proposals, specifications, and traceable requirements. The client always knows what they're paying for and when it will be delivered.
iv.
Honest scope.
Out-of-scope items are named explicitly upfront. Change requests are quoted separately. No surprise invoices, no scope creep on the engineering side.
03 / Products
Client-proven, productized.
Live · Demo on request
Clearance Studio
Trademark clearance screening across four independent axes - appearance, wordmark sight and sound, palette, device geometry - each with its own decision ceiling. Generated alternative marks are re-screened against every registration scored, and every published figure is pinned to a recorded run.
4 axesGenerated alternativesRun-stamped
In production · KSA
Document Extraction Engine
OCR plus LLM normalisation for invoices and forms in Arabic and English: 93.5 percent field accuracy on shipped MVP scope, a draft-approve-push review workflow, five ingestion channels including WhatsApp, nothing persisted by default.
AR / EN93.5%Review workflow
In production · Powers a KSA HR SaaS
Document Automation Engine
Branded bilingual document generation: Arabic and English templates with correct right-to-left rendering, server-rendered PDFs, per-tenant theming, no field values persisted. The engine behind a live platform running 34 templates.
34 templatesRTL PDFsNo persistence
Internal production · Pilots by arrangement
Private Knowledge Layer
Questions over your own files, answered with citations. Incremental scanning into Postgres and pgvector, sensitivity classification on every chunk, private sessions for private material, cloud calls off by default. Running daily inside the practice's own operations cockpit.
CitationsSensitivity gatingLocal-first
The product studio, in full → Clearance Studio → Document extraction →
04 / Selected Work
Recent work.
2026
Clearance StudioIP · Legal-Tech
Trademark conflict screening on four axes, each with its own decision ceiling, that then generates alternative marks which are re-screened against every registration scored.
PythonOpenCVNumPyFastAPIDocker
Context
A clearance search tells a client their mark conflicts with an earlier registration, and stops. What the client wanted was to file.
Problem
One blended similarity score buries the case that matters: an identical wordmark on completely different artwork averages down into the clear band, and that is exactly the mark an examiner refuses.
Approach
Four scored axes, each carrying its own ceiling that can decide the outcome on its own. On a conflict it generates alternative marks, re-screens each against every registration scored, and names every change rather than redrawing. Deterministic, so a score quoted to a client reproduces later.
Result
Live demoscreens a mark in about a second, access on request
2026
AI Invoice Parsing PlatformKSA · FinOps
Multi-tenant SaaS for automated invoice ingestion and field extraction, with hybrid OCR plus LLM normalisation and bilingual AR/EN.
PythonGeminiCloud VisionPostgreSQL
Context
Finance-operations platform for the Saudi market, bilingual Arabic and English, with WhatsApp document ingestion.
Problem
Finance teams re-key invoices by hand across formats and two languages.
Approach
Google Cloud Vision plus Gemini for hybrid OCR and field normalisation. Multi-tenant and stateless: documents stream through, nothing persists to disk.
Result
93.5%field-extraction accuracy on MVP scope
2026
Bilingual HR-Document SaaSKSA · HR-Tech
Multi-tenant, bilingual Arabic and English platform that generates HR documents on demand and streams them as PDFs, with live Saudi checkout.
Next.jsFirebaseGeideaPuppeteer
Context
An HR consultancy in the Saudi market needed to produce compliant HR paperwork across many document types, in Arabic and English.
Problem
Manual drafting is slow and inconsistent, and hard to keep bilingual and on-brand at scale.
Approach
A multi-tenant Next.js and Firebase platform with over thirty document templates, server-rendered to PDF and streamed to the browser with no field values persisted (PDPL-aware). Live payments through Geidea hosted checkout with Apple Pay.
Result
In productionbilingual AR/EN, taking real payments
2026
BC Auto Workshop SiteZA · Automotive
Migrated a locked-in GoDaddy site-builder presence to a self-hosted, hand-built WordPress site with structured-data SEO and a performance pass.
WordPressPHPCustom ThemeSEO
Context
A Cape Town auto workshop stuck on GoDaddy's Website Builder, which cannot export its own site.
Problem
The builder locked in the content, throttled performance and gave little control over on-page SEO.
Approach
Captured the legacy site, rebuilt it on self-hosted WordPress with a hand-built theme (no page-builder bloat), AutoRepair and FAQ structured data, 301-preserved URLs, and a local-SEO and speed pass.
Result
Off GoDaddyself-hosted WordPress, custom theme, technical SEO. View the site →
2026
Pinotage Villas Estate SiteZA · Property / HOA
Refreshed and hardened a residential estate (HOA) website on WordPress and Elementor, iterated against a local Docker mirror.
WordPressElementorPHPDocker
Context
A homeowners' association estate near Klapmuts in the Western Cape needed a cleaner, safer and better-performing site.
Problem
The existing theme was dated and fragile, with no safe way to test changes before they reached the live site.
Approach
Re-skinned and hardened the site on a vendored theme, iterated against a local Dockerised WordPress mirror, then deployed over FTP.
Result
Deliveredrefreshed estate site, delivered and fully paid
2025
Trademark Similarity EngineKSA · IP Rights
Multi-modal vector-search system that screens trademarks for visual and textual similarity, with a scraping pipeline and colour-distribution matching.
PythonCLIPpgvectorFastAPI
Context
Trademark clearance for registration in the Saudi IP market.
Problem
Manual clearance is slow and misses near-identical marks across visual and textual variation.
Approach
CLIP ViT embeddings for visual similarity, OCR plus a vision-language model for text, histogram and CIELAB colour, and fuzzy name matching, served from FastAPI over PostgreSQL with pgvector.
Result
94%top-5 visual-similarity accuracy; 87% combined visual and textual
2025
IoT Camera MiddlewareUS · Surveillance
Reverse-engineered the proprietary UDP protocol of low-cost IoT cameras to expose standard RTSP and MJPEG streams for off-the-shelf NVR systems.
GoProtocol RERTSPFFmpeg
Context
An integration layer letting cheap V720 and Yi-class cameras join standard surveillance stacks.
Problem
The cameras only speak a closed protocol through a poor first-party app, so they cannot join standard NVR systems.
Approach
Reverse-engineered the UDP protocol and built Go middleware that re-exposes the feeds as RTSP and MJPEG via FFmpeg.
Result
Open streamsstandard RTSP and MJPEG from a closed protocol
2024
Storefront PlatformDE · E-Commerce
Custom Shopify-alternative storefront with a visual page builder, featured collections and a merchant onboarding flow.
Vue.jsGoPostgreSQL
Context
A bespoke commerce platform for merchants who needed more control than hosted storefronts allow.
Problem
Off-the-shelf storefronts constrained layout and merchant onboarding.
Approach
A Vue.js front end with a visual page builder over Go services and PostgreSQL.
Result
Visual buildermerchant-configurable storefronts
Building something similar? Start a project →
- / The Practice
Led by an engineer who writes the code.
W-Enterprises is led by Wayne Fourie, a senior engineer with over a decade of production experience across AI/ML systems, full-stack web applications, and embedded middleware.
Recent engagements span AI-powered document parsing platforms, trademark similarity engines using CLIP embeddings and pgvector, reverse-engineered IoT camera middleware, and multi-tenant SaaS deployments for the Saudi Arabian HR-tech market.
The practice is small by design. The trade is between scale and rigour, and we choose rigour every time.
Wayne Fourie
Founder and Principal Engineer
- / How it works
How an engagement runs.
01 →
Discovery
We pin down the problem, the constraints and what success looks like. You get a senior read on scope and risk before anything is committed.
02 →
Spec and proposal
A written specification and PRD with traceable requirements. Scope is fixed and out-of-scope items are named, so both sides know exactly what ships and when.
03 →
Build
AI-accelerated delivery on serverless architectures, with monitoring, security review and CI in place from the start. Regular check-ins, no black box.
04
Ship and handover
The system lands in production with documentation and a clean handover. Code ownership and licensing are agreed in the proposal.
05 / Technology
The full toolkit.
Production-grade work calls for production-grade tools. The stack below reflects what we've shipped, not what's been touched in a tutorial. The specialised groups come first; the rest is one click away.
AI & Machine Learning
PyTorchTensorFlowLangChainLlamaIndexCLIPHugging FaceOpenAI APIClaude APIGoogle GeminiWhisperRAGVector SearchpgvectorChromaDBPydanticDocument AI
Specialized Domains
Multi-tenant SaaSOCR & Document AIComputer VisionVector SimilarityProtocol RERTSP / MJPEGFFmpegBilingual (AR/EN)RTL RenderingPuppeteer / PlaywrightWeb Scraping
Regional & Payments
Geidea (KSA)Moyasar (KSA)StripePayPalPDPL ComplianceZATCA-awareArabic LocalisationSARS / VAT
+ Show the full stack: languages, frontend, backend, data, cloud
Languages
PythonTypeScriptJavaScriptC#GoJavaPHPSQLHTMLCSSBashPowerShell
Frontend
Next.js 14ReactVue.jsNuxtReact NativeExpoElectronTailwind CSSshadcn/uiRadixZustandTanStack QueryRecharts
Backend & APIs
Node.jsExpressNestJS.NET / .NET CoreASP.NETFastAPIFlaskGoRESTGraphQLgRPCWebSockets
Data & Storage
PostgreSQLpgvectorSQL ServerMySQLMongoDBSQLiteRedisFirestoreSupabaseElasticsearchPineconeWeaviate
Cloud & Infrastructure
Google CloudFirebaseAWSAzureVercelCloudflareNetlifyDockerKubernetesTerraformGitHub ActionsGitLab CILinux
- / FAQ
Common questions.
What does W-Enterprises do?
W-Enterprises is a solo software consultancy that builds production AI/ML and full-stack systems end to end: RAG and LLM applications, computer vision, OCR and document AI, and multi-tenant SaaS. One senior engineer handles architecture through deployment.
Who is Wayne Fourie?
Wayne Fourie is the founder and principal engineer of W-Enterprises, a senior engineer with over a decade of production experience across AI/ML, full-stack web, and embedded systems, based in Cape Town, South Africa.
How can one engineer deliver platform-scale work?
By leveraging AI heavily across the build and working serverless-first, the practice delivers the output of a small team from one senior engineer, without agency overhead or handoffs. A typical platform goes from specification to production in around three weeks.
Do you build bilingual Arabic and English software and handle Gulf compliance?
Yes. Shipped work includes bilingual Arabic and English platforms with right-to-left interfaces for the Saudi and Gulf market, with PDPL-aware data handling and live Saudi payment integration through Geidea and Moyasar. Invoicing flows are built with Saudi ZATCA e-invoicing requirements in mind.
Which markets and timezones do you work in?
Remote-first from Cape Town at UTC+2, which overlaps the Gulf and Europe through the working day. Active engagements span South Africa, Saudi Arabia and the Gulf, Germany, and the United States.
How do engagements work?
Work begins with technical discovery and a written specification. Scope is fixed and documented, and out-of-scope items are named upfront. Code ownership and licensing are agreed in the proposal.
06 / Contact
Have a system that needs building?
The practice takes on a limited number of engagements concurrently. If your project is a strong fit, you'll have a response within one working day. Briefly: what you're building, when it needs to ship, and what success looks like.
Email
WhatsApp
Based
Cape Town
South Africa · UTC+2
South Africa · UTC+2
Operating in
ZA · KSA · DE · US
Remote-first
Remote-first
Availability
By engagement
Reply within 1 working day
Reply within 1 working day