Rainier PotgieterFull-Stack AI Architect

Agentic AI Engineering

Autonomous multi-agent pipelines and orchestration: continuous workflows that run without supervision across research, monitoring and intelligence gathering. This includes the unglamorous half of the discipline, instrumenting agents so their failures are visible, and evaluating non-deterministic systems that conventional software testing cannot cover.

Multi-agent orchestration, RAG, MCP, agent evaluation, MLflow tracing, LangGraph, Claude Agent SDK, LLM observability

Data Platform and Governance

Financial data lineage mapping every transaction back to its statement of record, built for BCBS 239 and IFRS 17 reporting. Decision-support systems built on a client’s existing data rather than a generic dashboard, answering specific operational questions: what to stock, where to cut, when to act.

Data lineage, BCBS 239, IFRS 17, audit trails, regulatory reporting, decision support, semantic discovery, PostgreSQL, Supabase

Application and Integration

Production applications, internal tooling and customer-facing products from architecture through to handover. Full-stack across modern TypeScript and Python, with a regulated-payments background that shapes how integration work is approached: compliance and audit requirements designed in from the start rather than retrofitted.

TypeScript, Python, Next.js, React, FastAPI, Node.js, Laravel, Flutter, Docker, payment gateway integration

Delivery Practice

Fixed-scope engagements with defined outcomes and measured results. Claims are backed by a pass rate, a coverage figure, a benchmark or a command output, not by assertion. Where an AI system is already in production and failing quietly, a structured reasoning audit identifies the failure modes and the concrete fixes.

Fixed scope and price, full IP handover, measured outcomes, reasoning audit, POPIA, production support

Governed agent tools with Unity Catalog and MCP, single and multi-agent systems, supervisor orchestration, MLflow tracing from prototype to production.

Verify at credential.net using reference dc762022

Knowledge Assistants, parsing unstructured documents with AI Functions, semantic chunking, Vector Search indexes, production knowledge assistants over multiple sources.

Verify at credential.net using reference 6f9c9fdf

Systematic evaluation of non-deterministic systems: built-in and guideline judges, custom judges, offline evaluation on curated datasets, online production monitoring, human feedback collection.

Verify at credential.net using reference e2f88f3c

Deployment via Declarative Automation Bundles, tool integration over MCP, MLflow Tracing instrumentation, production quality scoring with multi-turn judges and online evaluation.

Verify at credential.net using reference 1ba06019

Scope and pricing

Fixed scope and fixed price by default, with the deliverable defined before work starts. Retainer and embedded-contractor arrangements are available where a partner needs capacity inside an existing delivery team.

Intellectual property

The client owns the code, the infrastructure and the knowledge, in full, at handover. No retained components, no licensing tail, no dependency on me to keep the system running. Where I work under a partner’s master services agreement, that ownership passes through to the end client on the same terms.

Working under a prime contractor

I am comfortable being presented as part of a partner’s delivery team, working to their standards, their branding and their client-facing process. Direct client contact, or none, whichever the engagement requires.

Verification practice

Claims are backed by evidence rather than assertion: a pass rate, a coverage figure, a benchmark result or a command output. Where a system cannot be verified, I say so and name the evidence that would be needed. Verified means ready for review, not ready to ship. Human review, penetration testing or an audit pipeline belongs between the two.

Handover

Documentation, architecture notes and a walkthrough at the end of every engagement, written for the team that inherits the system rather than for the person who built it.

Compliance

POPIA compliance is baseline on South African engagements, with compliance implications flagged proactively wherever regulated data or operations are involved. GDPR-aware for United Kingdom work. Regulated-payments and financial reporting experience across BCBS 239, IFRS 17 and audit-trail generation.

Availability

Durban, South Africa, UTC+2. Working hours overlap the full South African and European business day and the morning of the United States eastern day. Remote by default, on site where an engagement requires it.