I design the systems that turn operational chaos into verifiable intelligence. Where information architecture meets behavioural psychology — that intersection is where most AI implementations quietly break, and where I work.
Location — Pretoria, SA · Remote contracts open Status — Available for new architecture engagements Focus — Human-AI workflow design
02 // The Problem
Why Most AI Deployments Fail
Most organisations deploying AI are solving the wrong problem. They treat it as a capability gap, when the actual gap is architectural. The model is not broken. The system feeding it is.
I came to this through a specific obsession: understanding not just how systems process information, but how humans produce it, distort it, and act on it. That intersection — between machine logic and human behaviour — is where most implementations quietly fail, and where I work.
Before AI, closing that gap required a team. Today it requires the right architecture. That is what I build.
03 // The Framework
Where Systems Actually Break
An AI model is only as competent as the workflow it enforces. My work sits at the intersection of three disciplines most organisations treat as separate — and that separation is exactly where the failure lives.
01 / Structure
Information Architecture
Imposing order on fragmented operational data before it enters any downstream system. If the input is not structured to be verifiable, no model output can be trusted.
02 / Human
Behavioural Psychology
Accounting for how humans actually produce and distort information. The workflows that survive contact with real teams are built around human behaviour, not against it.
03 / Tool
Workflow Automation & AI
Amplifying the output of both layers above. Automation is not the starting point — it is what becomes possible once the structure and human layer are sound.
04 // Deployed Work
Systems in Production
[CORE_PROTOCOL :: RUNNING]
Continuum — AI Progress Intelligence & The Hallucination Brake
The Business Problem: Stakeholders at Continuum, an early-stage development startup, were drowning in data but starving for the truth. Standard project updates showed activity but failed to prove progress — leadership could not distinguish between a workflow that was moving and one that was simply busy. The AI reporting layer had no way to know the difference either, so it reported confidently on a foundation no one had verified.
The Architectural Solution: A state-aware system designed to manage data entropy from the source. Rather than filtering bad data after the fact, the architecture intercepted it at the point of entry and structured it into verifiable truth anchors before any model consumed it.
01
Input Variables
Normalises raw administrative and structural system data — commits, state changes, logged hours — into verifiable truth anchors the system can reason against.
02
The Ghost Filter
Identifies tasks that simulate work without producing measurable output. Ephemeral activity that inflates velocity metrics gets flagged before it enters the reporting layer.
03
Efficiency Monitor
Tracks resource expenditure against concrete deliverables in real time. When time spent diverges from estimated hours, the system flags it immediately rather than surfacing it in a retrospective review.
04
Heartbeat Monitor
Detects task stagnation disguised as active progress. Work that has stopped moving but remains marked as in-progress gets surfaced to management automatically, regardless of task size.
05
The Hallucination Brake
The system's trust mechanism. A Global Confidence Score continuously evaluates the integrity of the underlying data and actively suppresses AI-generated summaries and projections when that data is weak, conflicting, or performative. The system does not just report on what is happening — it tells you how much to trust what it is reporting.
Currently mapping organisational bottlenecks and designing new human-AI workflows. Next system iteration pending deployment.
In Development
05 // Capabilities
What I Build With
These are not a skills list — they are the components of a repeatable method. Each one addresses a specific layer where human-AI workflows break down.
CAP-01
Process Architecture
Mapping and restructuring the sequence of operations that feeds your AI layer. Most automation fails because the underlying process was never designed for reliability.
Workflow MappingSystems DesignData Flow Analysis
CAP-02
AI Integration
Deploying LLMs within business systems in a way that accounts for the quality of inputs they consume. Model selection matters far less than context architecture.
Gemini APIPrompt EngineeringCloudflare Workers
CAP-03
Information Architecture
Structuring raw operational data into verifiable inputs. The difference between an AI output you can act on and one you cannot is almost always upstream of the model.
Designing systems that account for how humans actually behave inside processes — not how they are supposed to. The friction that kills adoption is almost always preventable.
Building the monitoring layer that tells you when a system is drifting before it fails. Confidence scoring, stagnation detection, and hallucination suppression.
Translating architectural decisions into working automations. The build layer is where the framework becomes something a team can actually use.
Make / n8nJavaScriptAPI Integration
06 // Diagnostics
Workflow Optimizer
Most workflow problems follow a pattern even when they feel unique. The symptoms vary — missed deadlines, unreliable reporting, AI outputs no one trusts — but the structural failure underneath is usually identifiable. Describe your bottleneck below in plain language. The system will return a structured diagnosis of the failure point and a recommended architectural fix.
This is not a chatbot. It is a diagnostic.
audit_protocol.exe
›
07 // Contact
Let's Look at the Architecture
If your team is moving faster than its processes, or you have deployed AI and cannot verify what it is telling you — that is the problem I am built for.
What happens when you reach out
01You describe the problem. One paragraph is enough. The structural failure is usually visible in how you describe the symptom.
02I send back a diagnostic. A short written assessment of what is actually broken and what fixing it would involve — before any commitment.
03We decide if it is worth building together. No sales calls, no proposal theatre. Either the problem is one I can solve, or I will tell you plainly.
