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
I built the infrastructure
South African AI was missing.
General-purpose AI models have no reliable access to South African statutory data. They approximate the estate duty abatement. They guess the prime lending rate. They cannot verify CIPC compliance dates or test SBC qualification against the five Section 12E conditions.
The SA Professional Financial Services MCP Server fixes that. 33 callable tools. 5 modules. 7 South African Acts fully encoded and maintained. Budget 2026/27 incorporated. Every CenturionAI platform calls this server when it needs a statutory calculation — and it gets the exact answer, with a legislation source and a tiered disclaimer.
# Estate duty calculation
$ mcp call calculate_estate_duty \
--gross_estate 4200000 \
--spousal_bequest 2800000
→ Estate duty payable: R0.00
→ Unused abatement: R2,385,000
→ S4(q) rollover applied: R2,800,000
→ Source: Estate Duty Act 45/1955 s4A
→ As at: Budget 2026/27
# The AI did not estimate. It called the tool.
Selected work
Each one came from a problem that was worked inside, not read about.
Insurance Comparison Platform
InsureLoansA
AI insurance comparison for South Africa: 9 product categories and 40+ providers, with the SAM assistant on WhatsApp, phone and web, confirmed Rand prices and no commission targets.
Document Intelligence Platform
SmartDoc AI
Production document analysis for South African contracts, deceased estates and CCMA matters: classification, red flags against SA norms, deadlines and plain-language summaries.
Voice AI Lead Reactivation
LeadRevive
Multi-tenant voice AI for lead reactivation: autonomous outbound calls, objection handling, Calendly booking and CRM logging, across WhatsApp and SMS as well as voice.
Multi-Agent Lead Engine
CenturionAI Proposal Generator
A four-agent CrewAI engine that produces a BRIDGE-framework business case with a 12-month Rand projection in about 90 seconds, with a LangSmith trace on every run.
Legal Research Assistant
Legal Eagle / Casewin
South African legal research: a six-agent pipeline with parallel search that returns a structured memorandum in SA citation format, with evaluator quality checks and matter memory.
Estate Administration Coordination
EstateHelp
Estate administration coordination across six phases: a client portal, a case dashboard and a 45-agent AI operating system, with statutory calculations from the MCP server.
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.
Try the diagnostic tool
Describe your business problem.
Get a board-ready business case.
Four AI agents analyse the problem, surface the real cost of inaction in Rand, design a specific technical solution, and compile a BRIDGE-framework business case. In 90 seconds. At no charge.
The proposal is emailed to both parties. When the discovery call happens, both people have already read the same document.
The problem
12-person accounting firm in Pretoria. Clients send contracts and tax documents. 3 hours per client manually reviewing. Cannot scale past 40 clients without hiring at R35,000/month.
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.
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.
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:
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