Platform systems, product delivery, and production-minded engineering.
End-to-end ownership: I design, build, deploy, and I'm the one who triages when things go sideways. AI-native platform engineer based in Somerset West near Cape Town, South Africa. I work across frontend, backend, AWS serverless systems, and delivery workflows that need to hold up once real users hit them, most of it sitting at the point where product configuration becomes operational reality. Currently at Inner Reality Limited.
I started coding in late 2022 alongside GPT-3.5: AI is native to my engineering process, not adopted later. I do not claim deep expert knowledge in every domain. I deliver production systems across all of them, because AI-augmented engineering collapses the gap between understanding and shipping. This site runs its own AI agent. Ask it anything about me.
The short version, without the filler
I'm a platform engineer with a production-first mindset and a strong bias toward systems that reduce friction without making operations worse. I'm most useful where a team needs someone who can move between interface, service, workflow, and infrastructure without losing the thread.
Platform-minded delivery
I work across the boundary between product and platform, co-architecting an AI-native multi-tenant SaaS alongside a 26-year-veteran CTO: schema design, system boundaries, blast-radius analysis, and the deploy gating that decides what is allowed anywhere near production. The point of all of it is that teams can launch, operate, and extend real features without brittle handoffs.
Rooted in production
I own production end to end: deploys, incident triage whenever it happens, and the post-mortem afterwards. The part that counts is what comes next: findings get codified into durable team rules, so a failure teaches everyone once instead of ambushing each of us in turn.
AI-native, harness-first
Intent-driven AI development is core to how I build: robust harness engineering and local small language models, not just hosted APIs. It is a discipline that barely existed five years ago, and it is real leverage for teams shipping under real constraints today.
From QA to platform architecture
Three years, three roles. A full-stack bootcamp first, graduated at 99% and Student of the Month, then quality assurance: the legacy system test suite in Python, and OpenSearch dashboards so stakeholders could finally see what campaigns were actually doing. Off my own back I built a Unity soccer AR prototype and wired it into the AWS platform; that is what earned me the move into development. Design, photography, and QA all fed the same instinct: make the thing clearer, reduce friction, build with intent.
Recent work that points to where I'm heading
A multi-tenant SaaS platform at Inner Reality, a solo-built esports platform, local AI tooling, and the site you're on right now. Different shapes, same centre of gravity: platform engineering, AWS delivery, and shipping things people actually use.
Campaign Creator
Primary author / Multi-tenant SaaS / AWS
Primary author of the no-code campaign platform at the centre of a multi-tenant SaaS for voucher redemption through gamification and augmented reality. Builder UX, template-driven campaign setup, asset handling and moderation, serverless submission and processing workflows, status tracking, notifications, and analytics entry points. It onboarded a large self-serve paying client on its own merits, and was cited as the deciding factor in closing a major enterprise deal. Product-facing on the surface, platform-shaped underneath.
Anti-cheat in experience delivery
Integrity design / Server-authoritative rewards
Gamified campaigns hand out vouchers and real prizes, so every gap between playing the experience and claiming the reward is an open invitation. I designed the barriers that sit in the delivery path: the reward decision is made server-side and treated as the only authority, a play is bound to its session and participant so a result cannot be replayed, forged, or passed around, and the routes that reach a claim without earning it are closed rather than discouraged. Removing the affordance beats asking people nicely.
WebAR and geofenced experiences
MindAR / WebAR / Turf.js + MapBox
A year on the platform's frontend, and the piece I would still point at: replacing the commercial Blippar AR SDK with open-source MindAR after evaluating four spatial engines, taking per-experience server cost to zero without losing the tracking quality campaigns depend on. Then GPS-triggered prize drop zones and geofencing with Turf.js and MapBox, so a reward can belong to a real place. A full data-capture form frontend shipped in a week, and a few hundred UI and UX defects across the interfaces went with it. Vue 3, TypeScript and Tailwind on the surface; MindAR and Three.js underneath.
WhatsApp and RCS chatbot integration
WhatsApp + RCS / Infobip / Conversational systems
Implemented conversational entry points into the SaaS platform on Infobip: session correlation, webhook routing, and capability exposure designed so a chat thread behaves like any other way into a campaign rather than a bolt-on beside it. RCS runs on the same routing model as WhatsApp, which is the whole point: a new channel becomes configuration instead of a second implementation.
Infrastructure and release discipline
CDK / OpenTofu / GitOps + SSM
The estate is defined in code: CDK, Lambda, API Gateway, DynamoDB, CloudFront, Cognito, S3, SSM for configuration and access, Valkey in front of the hot paths. Releases move GitOps-style through gates, with blast-radius analysis before anything reaches production and Playwright and Jest holding the line in the pipeline. I also sat alongside the CTO through a CDK-to-OpenTofu migration. How a live estate moves between infrastructure-as-code tools without a big-bang cutover is not something you pick up from documentation.
GG Anvil
Solo build / Discord-native / Textract OCR + RBAC
Founded and built solo, end to end, since January 2026: a Discord-native competitive gaming platform at goodgameanvil.com. CDK infrastructure, a Discord bot with role-based JWT auth and 25+ slash commands, a companion React portal, RBAC and billing, and one-click provisioning for a new org. Match scoreboards are read straight from screenshots by an async OCR pipeline: Textract for managed extraction, with fallback across Bedrock vision models and a local Tesseract adapter so one provider having a bad day never blocks a result. Around 220 commits in four months, and five P0 security findings found and closed. Solo does not mean unreviewed: least-privilege IAM, secure JWT handling, IDOR remediation and load testing are part of the build loop, not a hardening phase bolted on at the end.
Local AI Meeting Intelligence
Local ML / Sub-0.5s latency / Zero cloud calls
Three ML models running concurrently on one desktop GPU: Faster-Whisper for transcription, pyannote for speaker diarisation, and Gemma via Ollama summarising while the conversation is still happening. Sub-0.5s latency, under 5GB of VRAM, zero cloud calls. Nothing said in the room ever leaves the machine. Shipped as Bazzite Transcription, with emotion classification, topic detection, and a human-in-the-loop correction system that gets sharper with use.
This Is Luke (this site)
Bedrock agent / React 19 / CDK
The portfolio you are reading. An AI agent backed by AWS Bedrock represents me in the third person, rate-limited and prompt-injection-protected. React 19 and TypeScript on the front, Lambda and CDK underneath, built with the same intent-driven, AI-native process I bring to client work.
Engineering, but with the creative side left intact
Engineering is still a creative act
Architecture, decomposition, and judgment still matter. The tools move faster now, but the craft is still in how you shape the system, and in the unglamorous calls, like choosing the open-source engine over the licensed one because the cost curve is part of the architecture.
AI should expand the map
The real unlock is not hype. It is lowering the cost of trying good ideas, so engineers can think bigger and test directions that were previously too expensive.
Agentic governance is the premise
Underneath the AI tooling sits one idea: agentic governance baked into the operating system of the business, not bolted on after the fact. The same applies to people: a post-mortem that ends in a written rule is worth more than one that ends in a shrug. Get that right and the tools stop being magic and start being infrastructure.
Human intent stays central
These machines are useful because people bring taste, care, and responsibility to them. The technology is leverage, not the point.