Skip to main content
AI Engineer & FounderCenturionAICenturion, Gauteng

Built on the Industries
I Spent 25 Years Inside.

Most AI does something different every time it runs. That is fine for some problems. For sales pipelines, compliance checks, and document workflows — it is not. That is why everything built here starts with structure, not improvisation.

25+

Years of operational experience behind every build

R7.5M

Revenue growth delivered as a practitioner

3

Industries lived from the inside

0

Systems shipped that demo well and break in production

How the systems are built

Most AI is unpredictable
by design. Mine is not.

There is a version of AI that improvises every time it runs. It produces different outputs from identical inputs. That is fine for creative tools. It is not fine for a sales pipeline, a compliance workflow, or a client-facing system where consistency is the product.

The systems built here are deterministic wherever possible. Structured outputs. Typed schemas. Defined decision trees. The LLM is deployed where language understanding genuinely adds value — parsing documents, generating personalised communication, interpreting unstructured input. Everywhere else: logic, rules, and predictable architecture.

This is what 25 years of operational experience does to how you build software. You have seen what breaks. You build accordingly.

What that means in practice

Compliance checks

Run the same way for every document, every time. No variance in a regulatory context.

Lead scoring

The same input produces the same score. Not a different one depending on what the model felt like today.

Proposal generation

A defined structure the client can depend on — not an open-ended generation that surprises everyone.

Audit trails

Possible because the system's decisions are traceable. Not opaque.

LLMs are deployed where they win

Natural language understandingDocument interpretationDynamic communicationSemantic searchComplex reasoning over unstructured data

Selected work

Each one came from a problem that was worked inside, not read about.

The stack

Production-grade AI for South African businesses.

Python-based systems built to run when a client's business depends on them. Not prototypes handed over for someone else to finish. Every system shipped here carries the builder's accountability.

Agentic AI Systems

Multi-step workflows that observe, decide, and act. Built on LangGraph and CrewAI with human-in-the-loop oversight and fail-switch controls as standard.

RAG Pipelines

Retrieval-augmented generation over private document stores. ChromaDB, Pinecone, Supabase pgvector — structured retrieval, not guesswork.

Voice AI

Conversational voice agents via VAPI — for lead qualification, client engagement, and inbound handling that runs around the clock.

FastAPI Backends

Production-grade Python APIs. Multi-tenant architecture with row-level security. Built to hold up under real client load.

Multi-tenant SaaS

Full-stack platforms built on Supabase and PostgreSQL. Proper data isolation. Audit trails. Not MVPs that break at scale.

Compliance-first AI

FAIS, POPIA, NCA — not looked up before a proposal. The regulatory context is in the data model from day one.

Free tool · Live now4 AI Agents

Business case in 90 seconds.
What a consulting firm charges R50,000 to produce.

Describe your business problem. Four AI agents run a structured diagnostic — root cause analysis, solution architecture, industry proof points, and a compiled BRIDGE-framework business case with conservative Rand pricing.

Specific numbers. Named architecture. A three-phase build plan priced in Rand. Your actual industry context — not a generic template. In 90 seconds.

1
Business AnalystRoot cause + cost of inaction in Rand
2
Solutions ArchitectNamed stack + justified framework choice
3
Alter EgoIndustry-specific proof points
4
Proposal WriterBRIDGE document + three-phase build plan
Run the Diagnostic

No account required · Results in 90 seconds · PDF included · POPIA-compliant

CenturionAI BRIDGE Proposal

Financial Services · Sample output

Confidential

The problem

A 12-person accounting firm in Pretoria is capped at 40 clients. Manual contract review consumes 120 hours per month — the full capacity of one senior accountant at R35,000/month. Every new client added makes the bottleneck worse.

The solution

LangChain RAG pipeline on FastAPI. Contract ingestion, clause extraction, obligation tagging. Scales to 80 clients with the same team. POPIA-compliant data handling throughout.

Build estimate

R65K–R85K

Phase 1 build

4 months

To go-live

Month 5

Breakeven

Agents completed

Business Analyst

Solutions Architect

Alter Ego

Proposal Writer

How this works

One engineer. One product manager.
No hand-offs.

This is not an agency. There is no bench of contractors rotating in and out depending on the month. One engineer — me — designs and builds everything. That is a deliberate choice, not a resource constraint.

As the business grew, one role became impossible to ignore: someone needs to own the client relationship and the product feedback loop full-time. Not part-time. Not "when I come up for air between builds." Full-time.

Sanele is that person. CenturionAI's Product Manager — full-time, not background. Her job is everything that should not interrupt a build in progress. She is the reason this operation can take on serious client work without the founder becoming the bottleneck for every conversation.

The builder builds. The product manager manages. Neither role is compromised.

Two people·Full-stack delivery·Complete accountability

Sanele

Product Manager · Full-time

What she owns

  • Manages client communication from discovery call to delivered system
  • Runs the product feedback loop — what clients actually experience, not just what they report
  • Translates business requirements into something buildable before a line of code is written
  • Keeps project scope honest so what gets agreed is what gets delivered
  • Stays close to product decisions so the right things get built first

This model works for South African SMEs who need a technical partner who delivers — and who want the person who built the thing to be accountable for it.

Digital Twin

Don't read the CV.
Talk to the Digital Twin.

It has been trained on the full picture — not a polished summary. The career arc. The pivot decisions. The products and how they were built. What gets taken on and what gets declined. How the work is priced. How it is done.

Ask it directly. It answers the way I would.

Anthropic Claude+RAG+ChromaDB

Wandile — Digital Twin

Online

"Ask me about the products, the architecture decisions, the career arc, or whether this is the right engagement for your problem. I will answer the way Wandile would — directly, without the LinkedIn polish."

People usually start with:

How does the RAG pipeline handle compliance documents?
What makes your builds deterministic?
Are you available for a project in financial services?
Use the chat button ↘

Thinking in public

Writing from inside the build.

Start a conversation

Describe the problem.
The technology follows from that.

The most useful first conversation is never about which AI tool to use. It is about what is broken, slow, or expensive — and whether a well-built AI system can actually fix it. That question gets answered in 30 minutes.

Based in Centurion, Gauteng · Building for South African businesses — and wherever the work leads from here