Skip to main content
CenturionAI (Pty) LtdCenturion, GautengAI Engineering Studio

Agentic AI Systems for
South African Businesses.

Built to run. Not to demo.

CenturionAI designs and builds production-grade AI systems for financial services, legal, and business operations — by a team that has worked inside those industries, not just around them.

Why CenturionAI

Most AI consultants understand the technology.
We understand the industry it is being deployed into.

There is a version of AI consulting that works from the outside in — learn the tools, find a client, propose a solution. That version is everywhere right now. CenturionAI works from the inside out.

A decade inside South African financial services — at Momentum, Sanlam, Metropolitan, and Ithala Bank — holding the FAIS Key Individual licence at the peak of that chapter. Seven years running an SME logistics operation from the ground up. Those are not credentials listed to impress. They are the reason the AI systems built here are designed differently.

FAIS compliance is not looked up when it becomes relevant. It was managed daily, under FSCA accountability, for real. BCEA, POPIA, NCA, FICA — these are the operating context that shapes data architecture, audit trails, and access controls from day one.

When a voice AI system is built for a sales team, it is built by someone who made thousands of real sales calls, managed advisors, and knows what objection handling actually sounds like in the field. When a document analysis platform is built for a compliance department, it is built by someone who sat in front of an FSCA auditor and owned the answers.

That is what "built from the inside" means.

What Gets Built

Four categories of work. One standard of delivery.

Every engagement starts with a business problem. Not a technology selection. The tools are chosen after the problem is understood — not before.

01

Agentic AI Systems

Built on

Python · FastAPI · LangGraph · CrewAI · MCP · Railway · AWS

Multi-step AI workflows that take actions autonomously — research, analysis, classification, document processing, customer engagement, decision routing — without requiring manual intervention at every step.

Every system is built with human-in-the-loop oversight and fail-switch controls as standard. Not optional. Not an afterthought. Experience has shown that monitorability and the ability to intervene are not features to add later — they are what makes an AI system safe to run in a real business. Every build includes escalation logic, exception handling, and human override points by design.

Relevant for

  • Document-heavy workflows
  • Compliance-driven processes
  • Sales operations that need to scale without losing oversight
  • Financial services providers who need audit-ready systems with full traceability
02

Voice AI and Multi-Channel Engagement

Built on

VAPI · WhatsApp Business API · SMS · Supabase · Python · Calendly · CRM integrations

AI-powered engagement systems that reach leads and clients across the channels they actually use — voice calls, WhatsApp, SMS, email, and social media DMs — orchestrated from a single backend. Not a single-channel bot. A coordinated follow-up engine that adapts the channel to the context.

The sophistication is in the orchestration — knowing when to call, when to send a WhatsApp message, when to follow up by email, and when to hand off to a human. That logic is built into the system, not bolted on.

Relevant for

  • Insurance brokers and financial advisory practices
  • Real estate operations
  • Travel agencies
  • Spas, gyms, and appointment-driven businesses
  • Any sales-driven business with a lead database not being fully worked
03

Regulated Industry AI

Built on

FastAPI · Supabase RLS · Audit logging · Multi-tenant PostgreSQL · POPIA-compliant data handling

AI systems designed for environments where FAIS, POPIA, BCEA, NCA, and FICA are not optional considerations — they are hard requirements that shape every architecture decision.

Most AI developers treat compliance as a feature added after the system is built. The regulatory environment here is the starting point. Data models, access controls, audit trails, and output formats are designed around what a compliance officer, an FSCA auditor, or a POPIA review requires — because those environments were inhabited professionally for over a decade.

Relevant for

  • Licensed FSPs and insurance companies
  • Banks building internal AI tools
  • Legal practices with sensitive document workflows
  • HR and compliance departments automating regulated processes
04

Agency and Enterprise Partnerships

Built on

Full-stack AI delivery · PRINCE2 Agile methodology · SA regulatory context

For digital agencies, management consultancies, and enterprise clients who have the client relationships and the business context — but need a technical partner who can deliver agentic AI that holds up beyond the demo.

The partnership model is simple. Bring the client. Bring the brief. CenturionAI brings the technical architecture, the build, the deployment, and the domain knowledge that makes the system match what a regulated South African business actually needs.

CenturionAI does not do client-facing sales — the agency retains the client relationship throughout.

Relevant for

  • Digital marketing agencies whose clients are asking about AI
  • Management consultancies scoping AI transformation
  • Financial advisory networks
  • Enterprise procurement teams without in-house AI depth

Built With

Named tools. A deliberate architecture.

PythonFastAPITypeScriptNext.jsLangChainLangGraphCrewAIVAPISupabasePrismaPostgreSQLRAG architecturesVector databasesAWSRailwayVercelAnthropic Claude APIPayFastMCP tool architecturesMulti-tenant row-level securityWhatsApp Business API

Architecture philosophy

Deterministic where the problem allows it.

The no-code tools were tested first. Lovable, Make, and n8n were run under real conditions with real client requirements. They showed their limits. The move to Python-based agentic AI was a deliberate decision — not a trend. A builder's response to what actually holds up when a client's business depends on the system running.

Across all builds, the architecture is deterministic wherever the problem allows it. Structured outputs. Typed schemas. Defined decision trees. LLMs are deployed where language understanding genuinely adds value — document parsing, dynamic communication, semantic reasoning over unstructured data.

Everywhere else, the system behaves like engineered software: predictable, auditable, and consistent. The same compliance check runs the same way every time. That is not a limitation — it is what makes an AI system safe to run in a regulated environment.

Live Builds

These systems are deployed and running.
Not proposals.

SmartDoc AI

AI document analysis for the South African market. Three modules: business contracts, legal compliance, and estate administration. Under 60 seconds per document.

Pay-per-use from R199 · Security hardened · POPIA-compliant

FastAPINext.jsSupabaseClaudePayFast

LeadRevive

Multi-tenant voice AI platform for lead reactivation. Autonomous outbound sequences across voice, WhatsApp, SMS, and email. Qualification, objection handling, appointment scheduling.

Production · Multi-tenant · Multi-channel

VAPISupabasePythonWhatsApp APICalendly

Legal Eagle / Casewin

AI-powered legal research and document intelligence for South African practitioners and businesses. SA citation format, court hierarchy understood, secondary evaluator included.

SA citation format · POPIA-compliant · R199 per document

LangGraphFastAPINext.jsSupabaseClaude

INSURELOANSA

FAIS-regulated AI financial marketplace and funeral claims platform. Built on a decade of financial services compliance from the inside.

FAIS regulated · POPIA-compliant · Deterministic claims pipeline

VAPIFastAPISupabaseLangChainClaude

The Right Fit

Not every AI project is the right fit. Here is what is.

Regulated industries

Financial services, legal, insurance, credit — any environment where FAIS, POPIA, BCEA, or NCA compliance is non-negotiable. The depth here is not theoretical.

A real business problem with measurable cost

Time lost to manual document processing, leads not followed up, compliance processes that depend on human memory. If the cost can be quantified, the ROI can be demonstrated.

Agencies with client demand

If AI enquiries are arriving faster than in-house capacity can respond, a partnership conversation is worth having. CenturionAI delivers the build. The agency retains the client.

A South African operating context

FAIS, POPIA, the Master's office, the CCMA, PayFast, SA court case law, BCEA sectoral determinations. The daily operating environment, not reference material.

The Team

Small. Specific. Accountable.

Wandile Lokwe

Founder & AI Engineer

Architects and builds every system. 25 years of operational experience across customer services, financial services, and logistics — applied to AI engineering for the industries lived from the inside.

Sanele

Product Manager

Owns the client relationship from discovery through delivery. Manages communication, requirements, product feedback, and scope so builds stay on track and clients get consistent engagement throughout.

Carl

SAP Project Manager & Digital Transformation

PRINCE2 Agile-qualified. Brings structured project methodology and enterprise digital transformation experience to engagements requiring SAP integration, enterprise change management, or formal delivery frameworks. Primarily internal.

How It Works

Four steps. No ambiguity at any of them.

01

Discovery

A focused conversation about the business problem. Not the technology. What is breaking, what is slow, what is expensive, and what success looks like. If there is not a real fit, that is said directly.

02

Architecture and Scope

A clear technical proposal with defined deliverables, a realistic timeline, the specific tools and architecture, and how success gets measured. No vague scope. No open-ended retainers dressed up as strategy.

03

Build and Deploy

Python-based agentic AI systems built to production standards from day one. Security, multi-tenancy, audit trails, and compliance requirements are addressed in the build — not after it.

04

Handover and Support

Full documentation. Accessible codebase. A system the client's team can operate without needing CenturionAI in the room every time something needs to change. The goal is a functioning, maintainable system — not dependency.

Start Here

Describe what is broken.
The technology follows from that.

The most useful first conversation is never about which AI tool to use. It is about what the business needs to work better — faster, more accurately, at lower cost. The tools are a downstream decision.

wandile@centurionai.co.zacenturionai.co.zaCenturion, Gauteng, South Africa