Architect · Research Builder · Believer
Founder Director @ Kopano Labs · CPUT student · Cape Town, South Africa 🇿🇦

✝️ Jesus is King.
I want the things I build to serve people before they impress people.

I build software, AI systems, games, governance protocols, financial agents, robotics experiments and strange little research machines from Cape Town.

But the thread underneath all of it is becoming clearer:

How do we give intelligent systems more capability without giving up human authority, identity, accountability, memory or truth?

I started by building applications. Then I became obsessed with what happens when software remembers, acts, coordinates, changes the world, and has to explain why it did so.

That is where most of my work lives now.

I care about people who do not have perfect infrastructure, perfect bandwidth, perfect devices, perfect credentials or time to decode a complicated interface. South African realities — unemployment, transport cost, township infrastructure, education, small business, safety and opportunity — are not “edge cases” in my work. They are design inputs.

The current research direction is moving beyond “which AI model is smartest?”

MODEL      = capability
INTERFACE  = embodiment
SEAT       = authority
IDENTITY   = accountability
GOVERNANCE = continuity

A model can change. A UI can change. A vendor can change.

The harder question is whether identity, authority, context, evidence and responsibility survive the change.

That is why I increasingly treat models as swappable capability layers rather than the owner of the system.

PERSISTENCE × CONSISTENCY × CONTEXT
                ↓
       GOVERNED CONTINUITY

This is the research spine connecting KPGS, Project Jennifer, RUNE, Aya, LEFA, Digital Hippocampus, PKA, MMAO/MAO and the wider Kopano Labs estate.

You do not need to understand my architecture vocabulary to enter the work.

_{The telemetry card tracks a selected core product estate. The research radar above is intentionally broader.}

Project Jennifer has evolved hard.

It is now centered on Jennifer City: a web-first companion-consequence RPG about relationships, memory and governed intelligence. Tactical combat remains a long-term direction; the present product proof is a loved continuity loop.

CHOICE
  ↓
CONSEQUENCE
  ↓
RECEIPT
  ↓
PERSISTENCE
  ↓
THE NEXT SCENE REMEMBERS

The current repository proof now includes a CI-gated love loop, playable Memory District / Telemetry Tower work, a Third Signal episode, local-first continuity, epistemic disagreement, and a same-browser Continue path designed to survive API failure.

That matters to me because I do not want AI characters that only sound like they remember you. I want the system to distinguish story, memory, inference, authority and evidence.

RUNE — Runtime Unified Network Endorsement is my current public research product for a very specific problem:

Multi-agent coordination is not the same thing as verified safety.

RUNE treats agent coordination events and identity claims as things that can be independently endorsed, rejected, revoked and receipted — rather than trusting agreement because several agents voted the same way.

The public repo is intentionally honest about status: reference MVP, scaffold + fail-closed gate. It is not marked complete, proven or demo-ready without owner verification receipts.

MCP  → agent ↔ software tools
MHS  → agent ↔ physical hardware
RUNE → agent ↔ agent coordination / identity claims
KPGS → wider governance authority + continuity

The more I build, the more I care about heavy architecture disappearing behind a humane interface.

Aya is a local-first multimodal design-thinking desktop companion for students and makers.

The MVP is deliberately narrow: speak, type, sketch or show an intention → clarify missing requirements → create an explainable brief → make one approved Canva change and one approved SolidWorks change.

The point is not “AI does CAD for you.” The point is a learner staying in control while the system helps them think, make and understand.

LEFA asks a similar question in finance:

Can one human-facing intelligence receive messy intent, use governed internal agents and real financial evidence, preserve uncertainty, and make a decision that time can later validate?

Its current lane includes zero-trust middleware, bounded paper-execution authority, evidence-backed chains, ADK work, risk gates and receipts. Live trading is not silently promoted from a demo.

A lot of my repositories are separate products, but the research is converging.

HUMAN INTENT
    ↓
IDENTITY ── who is acting?
    ↓
SEAT ────── what authority do they hold?
    ↓
CONTEXT ─── what should persist?
    ↓
GOVERNANCE ─ what may change?
    ↓
EVIDENCE ─── what proves it?
    ↓
RECEIPT ──── what survives the session?
    ↓
ACTION / WORLD CHANGE
    ↺

Current research lanes include:

• KPGS / GSMB — governance continuity across stateless model sessions.
• Digital Hippocampus — persistent context, provenance and memory architecture.
• PKA / POCvsFOC — bounded uncertainty and an immune system for false closure.
• MMAO / MAO — multi-agent orchestration with human authority held outside agent consensus.
• RUNE — independently endorsed coordination and identity claims.
• Memory Receipts — making consequence inspectable in Project Jennifer.
• Source Authority — relevance is not the same as permission, canon or truth.
• Zero-trust agent systems — fail closed, minimize authority, preserve evidence.
• MCP / WebMCP / ADK / portable skills — reusable interfaces for agents without surrendering governance to one vendor.
• Embodied and spatial computing — Three.js, Blender, robotics, voice interfaces and physical-tool control.

POC BEFORE NARRATIVE
RECEIPTS BEFORE CLAIMS
SOURCE AUTHORITY BEFORE RETRIEVAL CONFIDENCE
GOVERNANCE BEFORE DRIFT
UNKNOWN IS A VALID STATE
FAILURE MUST LEAVE A RECEIPT
MODEL CAPABILITY DOES NOT EQUAL SYSTEM AUTHORITY
CONSENSUS DOES NOT EQUAL ENDORSEMENT

I am building from South Africa, not from an abstract “ideal user” diagram.

I think about:

• the student with an idea but no expensive workstation;
• the graduate with skills but no clear opportunity path;
• the township business that cannot afford wasted transport or data;
• the founder who needs software to tell the truth when a provider is down;
• the designer who wants AI help without surrendering authorship;
• the person using financial intelligence who deserves boundaries, not magic language;
• the developer who needs to inspect what the agent actually did.

That is why I keep coming back to offline-first, low-data, local-first, governed, reversible and evidence-backed systems.

Technology should increase agency.

I use OpenAI, Anthropic, Google, Microsoft, AWS, Hugging Face and open-source models/tools where they fit.

I do not want vendor lock to become architecture. The model is a capability provider; the system still needs its own identity, authority, memory and governance.

I am still a student while building all of this.

Some days I am writing governance protocols. Some days I am debugging a rover. Some days I am learning Blender from the beginning, studying algorithms, reading about Fourier optics and holography, or trying to understand why a CI gate exploded at 20:00 on a Sunday. 😭

I like that tension.

I do not want to become the kind of builder who can explain a system beautifully but has stopped being teachable.

If you are a student, researcher, founder, designer, developer, investor, community builder or simply curious human, you do not need the perfect introduction.

If something here makes you think “wait… what are you doing over there?” — that is a good enough reason to talk. 🙂

Build for people. Tell the truth. Leave receipts.
From Cape Town outward.
_{Living profile · major evolution: September 2026}