Available · Johannesburg · Remote
I build AI systems
for regulated work.
I’m an applied AI engineer building practical tools for a licensed funeral assurer in Johannesburg. My work includes a live lead-management CRM, digital applications and document workflows. Available for remote work.
Live demo
Ask the assistant.
Grounded in my CV and public repos. Refuses out-of-scope questions and cites sources. Source code
Embed
query to vector
Retrieve
cosine top 6
Gate
score ≥ 0.35
Generate
grounded
Cite
sources
Ask about my projects, stack, or availability:
Per-instance cap · grounded · refuses out-of-scope · email Mike
Work
Selected projects.
Shipped systems, a live AI demo and an approved workflow, followed by more work.
Featured work
Regulated insurance intake: multi-step form, server-side validation, policy PDF generation, and staff admin — for funeral-assurance workflows at Zororo-Phumulani.
Built digital intake and document generation for funeral-assurance applications.
Python · Flask · SQLite · Docker
Read case studyProduction CRM at Zororo-Phumulani Funeral Assurance: used by over 40 people across around 10 roles.
Built lead routing and access controls for a CRM used by over 40 people; live since approximately July 2026.
Python · FastAPI · PostgreSQL · Redis
Read case studyA retrieval-augmented assistant with grounded answers and clickable source citations.
Built this live assistant to demonstrate grounded retrieval and citations.
Next.js · TypeScript · Gemini embeddings · Cosine retrieval
Read case studyEvidence-grade e-signatures and a requester → manager → Accounts approval workflow, built as the second domain of Zororo Document Intelligence.
Designed a signature and approval workflow with tamper-evident records; not yet live.
Python · FastAPI · PostgreSQL · Alembic
Read case studyMore work
Governed AI control plane: every event passes Presidio PII detection, an OPA policy check, risk scoring, then ALLOW / MONITOR / APPROVE / BLOCK, with a hash-chained audit log.
Python 3.12 · FastAPI · Presidio · OPA/Rego
Read case studyRole-based platform for funeral-case processing: claims, cash services and courtesy services.
Next.js · TypeScript · NextAuth · Prisma
Read case studyThree agents (research, strategy, content) turn signals from GitHub, ArXiv and O'Reilly into platform-ready posts.
Python · FastAPI · Jinja2 · Ollama
Read case studyGoverned AI operations platform (prototype): LangGraph state graph with human approval gates, Postgres checkpoints, risk-tiered tool gateway, and pgvector hybrid retrieval with citations.
Python · FastAPI · LangGraph · PostgreSQL
Read case studyCoding agent (prototype): own tool-calling loop, provider-agnostic adapters, approval prompts before writes, sandboxed eval harness.
Python · Anthropic Claude API · OpenAI API · pytest
Read case studyClaude vision extraction through Pydantic schemas, per-field confidence scoring, and a needs-review flag for human verification.
Python · Anthropic Claude API · Pydantic · FastAPI
Read case studyMulti-tenant SOC platform: event ingestion, MITRE ATT&CK correlation, automated response, and POPIA-aware data handling.
Python 3.11 · Flask · PostgreSQL 15 · Redis 7
Read case studyReference work and labs at github.com/MikeNcube.
About
Mike S Ncube.
Applied AI Engineer and Systems Architect at Zororo-Phumulani Funeral Assurance (FSP 48558), where I have worked in AI engineering since June 2024. I built Digi App Form, the lead-management CRM used by over 40 people, and ZDI (approved for production rollout) in a four-person IT team.
At Zororo-Phumulani, I moved from call-centre work into frontend development in 2022, then into AI engineering in 2024. Explore the live assistant above, the project case studies, and my public GitHub portfolio.
How I build
- Models are fallible dependencies
- Validation, retries, and a fallback path on every LLM call.
- Every run is auditable
- Structured logs, trace IDs, audit entries for state changes.
- Pipelines are contracts
- Pydantic schemas at boundaries, validation before persistence.
- Labels are honest
- Production, prototype, or lab — each repo says what it is.
Training
- AWS Becoming an AI Practitioner Series — Sep 2026
- Retrieval-Augmented Generation — DeepLearning.AI
- Generative AI with LLMs — DeepLearning.AI
- Agentic AI — DeepLearning.AI
Contact
Get in touch.
Available for full-time or contract roles, remote from Johannesburg. I respond within 24 hours.
Johannesburg · Remote-first