Technical skills
Three tiers, defined strictly, with the evidence attached.
There are no percentage bars on this page. Nobody can say what "Python 72%" means, and a skills list that implies that kind of precision is the clearest sign of an inflated portfolio. Instead every item sits in one of three tiers and states what is behind it.
Demonstrated 23
There is a working artefact or a verifiable qualification behind this. Ask me to show it and I will open it, not describe it.
Working knowledge 14
Used successfully in shipped work, with AI assistance. I can explain my decisions and defend them. I am not claiming unaided fluency.
Currently learning 5
Under active study. Listed for direction, not as a capability claim. Nothing moves out of this tier until there is something built with it.
Deliberately absent: AWS, Kubernetes, Docker, TypeScript, React and model training appear nowhere on this page, in any tier. I cannot demonstrate them, so they are not skills — they are targets, and they live on the career roadmap instead. Python appears only under working knowledge, with an explicit note, because I have shipped real Python with AI assistance and have not completed formal fundamentals.
Category 1 of 6
AI and Generative AI
Where most of my practical work sits. Every item here is backed by a system described on the Projects pages.
7 demonstrated 2 working knowledge 2 learning
-
Applied generative AI Demonstrated
Two shipped desktop applications built on generative speech and language models, running locally on the user device.
-
LLM application design Demonstrated
A spoiler-safe companion enforced by context windowing rather than prompt instruction, and an autonomous presenter whose director loop treats the model as one input among several.
-
AI agents and autonomous loops Demonstrated
An unattended presenter that decides what to respond to, when to speak and when to change topic, with no operator input during a run.
-
AI content evaluation Demonstrated
Automated narration verification by speech recognition as a release gate, plus the review practice that caught a factually correct chart teaching the opposite of its lesson.
-
AI workflow design and automation Demonstrated
Multi-stage production pipelines with gates: twenty animated episodes produced from structured data, unattended, with an automated check that can refuse to deliver.
-
AI-assisted software development Demonstrated
The working method behind everything on this site: specify, direct, test, verify, reject. Described without euphemism on the About page.
-
Speech synthesis and recognition (applied) Demonstrated
Per-character voice assignment with episode-long consistency, read-ahead synthesis scheduling for zero-latency output, and speech recognition used as a verification instrument.
-
Prompt design and output constraint Working knowledge
Register filtering to stop a model sounding like an assistant; personality expressed as a constraint on style rather than as an instruction.
-
Model evaluation methodology Working knowledge
Split-half validation, plateau-over-peak parameter selection and reproducibility control, applied rigorously to a simulation-driven system. The transfer to ML evaluation is real, but I have not yet applied it to a model.
-
Retrieval-augmented generation Currently learning
Not built. Listed honestly as a gap: I have built context-constrained LLM features but never a retrieval pipeline over a vector store.
-
Fine-tuning and model training Currently learning
Not built. I have run and integrated models, never trained one.
Category 2 of 6
Testing, Evaluation and Quality
The strongest part of my profile, and the part most relevant to AI quality and evaluation work. Each item is a practice I can show the evidence for, including the times it caught me out.
7 demonstrated 1 working knowledge 1 learning
-
Experimental design and validation Demonstrated
Split-half in-sample and out-of-sample adoption testing, with changes rejected for failing one half despite improving the total.
-
Measurement integrity Demonstrated
Discarded a full session of my own results after finding the simulator setting invalidated them; established that a platform's own success metric was wrong for my system and rebuilt the measurement.
-
Automated verification gates Demonstrated
Delivery gates that measure resolution, frame rate and audio level and roll back on failure; speech-recognition verification of narration; an automated information-boundary test; and the build gate on this site.
-
Content and factual review Demonstrated
Instructional material fact-checked under a compute-before-you-write rule, which caught both a materially wrong figure and a chart that was accurate and misleading by construction.
-
Testing against reality, not fixtures Demonstrated
A rule affecting real people passed its fixture and produced false positives against the most legitimate users when run on real data. Fixtures are now treated as necessary and insufficient.
-
Reproducibility discipline Demonstrated
Seeded determinism through a single generator after finding correlated seeds; fixed execution settings so simulation runs are comparable at all.
-
Regression and rollback practice Demonstrated
Checksum-stamped builds with retained rollback artefacts, and a delivery gate that restores the previous output rather than shipping a failed one.
-
Adversarial testing Working knowledge
Adversarial questioning of a spoiler-constrained companion to verify the context boundary holds under pressure.
-
Structured annotation and rubric design Currently learning
Not yet done formally. My quality judgements are documented and consistent but have not been expressed as a scored rubric another person could be calibrated against.
Category 3 of 6
Information Technology
Formally qualified here, and it is the layer everything else runs on.
4 demonstrated 2 working knowledge
-
IT support fundamentals Demonstrated
Higher Certificate in Information Technology Support Services, IIE Rosebank College, completed March 2024. Hardware, operating systems, networking fundamentals and user support.
-
Systematic troubleshooting Demonstrated
Diagnosed a repeating GPU-level display failure on a render workstation to a verified root cause; traced delivered-output defects back through an automated geometry scan rather than by guesswork.
-
Technical documentation Demonstrated
Architecture notes, deployment runbooks and handover documentation — including this site's own README, deployment guide and the commented rationale in its source.
-
Software testing Demonstrated
See the Testing, Evaluation and Quality category above.
-
Windows workstation and GPU operations Working knowledge
Running a GPU render workstation: driver-level fault diagnosis, storage lifecycle under multi-hundred-gigabyte workloads, and long unattended job scheduling.
-
Systems thinking Working knowledge
Pipeline architectures with explicit stage boundaries, and an engine separated from its presentation layer so the simulation can be tested without a renderer.
Category 4 of 6
Information Security
Academically my strongest subject (distinction at NQF level 8), and applied in product decisions rather than in a security role.
2 demonstrated 2 working knowledge 1 learning
-
Information security fundamentals Demonstrated
INF4831 Information Security, University of South Africa, NQF level 8, passed with distinction (2021).
-
Privacy-by-design decisions Demonstrated
A shipped application that processes entirely on-device, requires no account, stores no video and transmits nothing — an architecture decision, not a policy page.
-
Credential and secret handling Working knowledge
OS-level encrypted storage for user-supplied keys; and a pre-publication audit of this repository that found and removed server credentials before anything was made public, including rewriting them out of git history.
-
Risk assessment concepts Working knowledge
Applied to a copyleft dependency recorded as a condition on commercial use, and to deciding what may and may not be disclosed about a system that is still running.
-
Security operations and incident response Currently learning
Not done. No SOC, SIEM or incident-response experience of any kind.
Category 5 of 6
Product and Delivery
Taking something from an idea to a released, documented artefact that other people can actually use.
3 demonstrated 1 working knowledge
-
Digital product development Demonstrated
Two released Windows applications taken from concept to a public download, and a production pipeline that delivered twenty finished episodes.
-
Honest product communication Demonstrated
Stating the hardware requirement up front rather than after installation, and publishing a "not built, or not proven" list for every project on this site.
-
Technical SEO and structured data Demonstrated
This site: JSON-LD for Person, CreativeWork and BreadcrumbList, canonical URLs, Open Graph, a generated sitemap and robots.txt, with title and description lengths verified by an automated check.
-
Product research and requirements Working knowledge
Defining a product by its constraints first — the four things one application must never do eliminated most architectures before any code existed.
Category 6 of 6
Engineering and Tooling
Stated carefully. These are languages and tools I have shipped real work in, with AI assistance, and that is the claim — nothing more.
6 working knowledge 1 learning
-
JavaScript / Node.js Working knowledge
A dependency-free simulation engine with 38 tests, three Electron applications, render orchestration, and the zero-dependency generator that builds this website.
-
HTML, CSS and accessible markup Working knowledge
This site, built from scratch with no framework: semantic structure, keyboard navigation, visible focus, reduced-motion support and a print stylesheet.
-
Python Working knowledge
Production pipelines: episode builders, render queues, 3D engine automation and analysis scripts. Written with AI assistance; formal fundamentals are on the roadmap and not yet complete, which is why this is not listed as a professional competency.
-
Electron desktop applications Working knowledge
Three applications, two of them released to the public.
-
ffmpeg and media processing Working knowledge
Assembly, mixing, loudness handling and container verification across twenty delivered 4K episodes.
-
3D production tooling Working knowledge
Scripted set construction, character import, rigging and retargeting, and cinematic camera work in Unreal Engine and Blender via their Python APIs.
-
Git version control Currently learning
I commit locally and use history, but I have not worked in a branching team workflow, reviewed a pull request or resolved a shared-branch conflict. Stated as learning for that reason.
How to test any of this
Every "Demonstrated" item above points at a system on this site. I would rather open one and walk through the code and the test output than describe it. A claim that does not survive that should not be on this page.