Data→Intelligence→Action→Value
A practical blueprint for turning enterprise data and AI into measurable operational performance, productivity and growth.
Everything below runs in your browser on synthetic data. No Sasol confidential or proprietary data is used, and the AI never writes to a control system.
Short on time? Start here.
Former Software Developer at Securities Trading Technology (STT). 3+ years building mission-critical clearing, trading and client-analytics platforms in regulated financial services. Based in Johannesburg.
A working industrial AI lab: three machine-learning models that detect, explain and optimise, a second use case on the same platform, and a knowledge assistant that cites its sources.
Value first, trusted data, one reusable platform, responsible AI and real adoption. Understand in 90 days, prove value by 180, scale by 365.
The AI never writes to a control system. It advises and an accountable person decides. When the data or the model can't be trusted, it steps back and escalates.
I haven't yet led an enterprise AI function. This page is labelled throughout so you can see what I have done, what transfers, and what I propose.
I don't just propose AI.I show how it becomes an enterprise capability.
Clearing and trading platforms: microservices, secure APIs, messaging and cloud delivery in regulated financial services.
Start from the decision and the KPI, then choose the data and the model. Value first, technology second.
Turn one working use case into shared pipelines, lifecycle, governance and measurement that every team reuses.
Sasol AI Operations Lab
Watch three real machine-learning models spot a problem on a simulated process unit, explain it and suggest what to do. A person always decides.
- Synthetic data only, no Sasol data
- Never writes to a control system
- Runs entirely in your browser
One platform, reused.
Scale comes from reuse. Two more working examples built on the same building blocks as the Lab.
Same platform. A second use case.
Answers from approved sources, with citations, or no answer at all.
Ask about Sasol's public strategy, emissions, operations or digital history. The assistant retrieves from a small, approved set of public sources and cites every statement. If the answer isn't there, it says so.
Ask the knowledge assistant
28 notes · 9 public sources · as of 30 Sep 2026Pick a question below or type your own. Try something it can't know too, like today's share price.
How it stays trustworthy
- Only sources classified PUBLIC are indexed. Confidential or personal questions are refused.
- Notes are short paraphrases with a link to the original, never copied report text.
- Every statement in an answer carries a citation you can open.
- Below a relevance threshold the assistant declines instead of guessing.
- In production: embeddings and a vector store behind the AI gateway, access control per user, and an LLM that writes the answer under this same contract.
Approved sources
- Climate FAQs · Sasol, 2025
- Capital Markets Day 2025 — CEO and CFO script · Sasol, May 2025
- FY25 annual results summary · Sasol, Aug 2025
- 2025 Integrated Report (overview) · Sustainability Reports, 2025
- Sasol progresses CMD commitments · Engineering News, 3 Oct 2025
- Sasol and Air Liquide renewable energy update · Engineering News, 31 Oct 2025
- Sasol deploys technology to pinpoint emissions-reduction opportunities · Sasol media release, Aug 2022
- Sasol to use Honeywell Connected Plant · Honeywell press release, Jul 2018
- Sne Dlamini named group chief information and digital officer at Sasol · CIO South Africa, Jun 2026
From one demo to an enterprise capability.
My proposed approach, one topic at a time. Pick the one you care about most.
A prototype is not a production system. Here is the difference.
The Lab collapses the whole chain into a browser. In an enterprise, every arrow becomes a governed, owned and monitored capability.
- 01Industrial systemsDCS · historian · LIMS · CMMS→
- 02Governed data platformingestion · quality · lineage · data products→
- 03Enterprise AI platformfeature store · training · registry · serving · monitoring→
- 04AI models / agentsversioned · validated · approved→
- 05Operational applicationscontrol-room UI · work management · mobile→
- 06Human decisionsoperator · engineer · manager→
- 07Measured outcomesbaselined KPIs · value ledger
Detection, explanation, policy gating and human approval carry over unchanged. What changes is who owns each step and how it is secured, monitored and supported.
Most of the effort in production sits in data pipelines, integration with operational systems, change management and value measurement.
Recommendations flow to people through operational applications. The AI platform has no write path to the DCS.
One use case.One reusable enterprise pattern.
- 1Single AI use case
- ↺Reusable data pipeline
- ↺Reusable model lifecycle
- ↺Reusable governance
- ↺Reusable AI gateway
- ↺Reusable monitoring
- ↺Reusable value measurement
- ∞Enterprise AI platform
The value is not only in the model. The value is in creating a repeatable capability for deploying many models safely and economically.
The same pattern, applied across the enterprise.
Select a domain. Each reuses the data, lifecycle, governance and measurement built for the Operations Lab — only the problem, data and decision change.
- Business problem
- Units drift from their best operating point; the loss is noticed late, if at all.
- Data
- Historian telemetry, lab quality results, operating targets, shift logs.
- AI technique
- Anomaly detection, surrogate models, constrained optimisation — the Lab pattern.
- Human decision
- Shift supervisor and process engineer approve setpoint changes.
- KPI
- Energy intensity (GJ/t), yield, off-spec hours.
- Governance
- Medium risk. Advisory only, human approval, drift monitoring.
- Scale pathway
- One unit → a site's similar units → a template per unit type.
Every stage has an exit criterion. That is how pilots stop piling up.
Ideas enter as business problems. Only those that pass each gate consume more investment. Counts shown are illustrative.
Is there a measurable business outcome?
Baseline and target KPI agreed with finance.
The Operations Lab walked through these gates in miniature: a framed problem with a KPI, a data-quality gate, validated models, human-approved decisions, a value estimate, and lifecycle controls ready for industrialisation.
Trusted data is the product. AI is one of its consumers.
Select a stage or a cross-cutting control. The controls span the whole architecture; they are not a final checkpoint.
Owned, documented and quality-scored datasets with contracts — the unit of reuse across the enterprise.
Enterprise SSO (SAML/OIDC) — every request tied to a named identity
Viewer · operator · engineer · model owner; approve rights by role
Evidence packs tagged; only approved classes may leave the boundary
Copilot tools are read-only; no write path to control systems
Every inference, recommendation and human decision recorded
API keys in a vault / server env — never in the browser
Gateway: authN, rate limits, payload limits, schema validation
Only approved, versioned models are callable in production
No operating change without an accountable person's approval
No credentials are included in this application. The optional LLM route reads its key from a server environment variable only.
Trust is designed in, one control at a time.
AI recommends and people decide. Accountability for an action always sits with a named person.
An enablement engine, not a central bottleneck.
The CoE sets standards and builds shared capability. Business teams build on it. Governance stays consistent because it lives in the platform, not in a queue.
- 1CoE creates standards↓
- 2Business teams build↓
- 3Reusable capabilities are shared↓
- 4Governance remains consistent↓
- 5Value is measured↓
- 6Successful patterns scale↺
AI transformation is a people transformation.
Uses approved AI tools every day and knows when not to trust an output.
Fictional users. Each business team has champions (gold) who connect their colleagues to a shared community of practice and to each other.
From efficiency to growth.
Cost and productivity fund the journey. Better decisions compound it. New data products and digital services change what the business can sell.
- Run better
- Cost · Productivity · Reliability
- Decide better
- Forecasting · Optimisation · Intelligence
- Grow differently
- Data products · Digital services · New business models
Verified, batch-level carbon-intensity data offered to industrial customers who need it for their own reporting.
Illustrative future opportunity · not a Sasol initiativePredictive delivery and availability insight offered to key customers as a digital service.
Illustrative future opportunity · not a Sasol initiativeOne optimisation across production plan, utilities and tariffs, rather than separate plans.
Illustrative future opportunity · not a Sasol initiativeFast what-if analysis of feedstock, price and operating choices for planners.
Illustrative future opportunity · not a Sasol initiativeGoverned retrieval over procedures and incident learnings, with citations, for control-room staff.
Illustrative future opportunity · not a Sasol initiativeCo-developed solutions with technology partners and universities, with a clear IP and data-sharing model.
Illustrative future opportunity · not a Sasol initiativeManage AI as a portfolio of investments, not a list of projects.
Every initiative carries investment, expected value, realised value, adoption, risk and stage — so leaders can scale what works and stop what doesn't.
Bubble size = adoption. Colour = stage: grey discover · dark green pilot · green scale · gold industrialise.
Where the value gap sits: not started, piloting, adoption or delivery.
Fictional monthly active users (green, left) and champions (gold, right).
Discover → Pilot → Scale → Industrialise.
Top-left: quick wins. Top-right: strategic bets needing stronger governance.
| Initiative | Area | Stage | Risk | Value type | Invest (R m) | Expected (R m/yr) | Realised | Adoption |
|---|---|---|---|---|---|---|---|---|
| Process energy optimisation | Operations | Pilot | Medium | Cost | 6 | 28 | 4 | 42% |
| Rotating-equipment health | Asset management | Scale | Medium | Productivity | 9 | 35 | 17 | 64% |
| Utility dispatch optimiser | Energy | Discover | Medium | Cost | 2 | 22 | 0 | 0% |
| Demand sensing | Supply chain | Industrialise | Low | Productivity | 7 | 18 | 15 | 81% |
| Logistics route optimiser | Supply chain | Scale | Low | Cost | 4 | 12 | 7 | 58% |
| Key-account intelligence | Commercial | Pilot | Medium | Revenue | 3 | 15 | 2 | 35% |
| Price-elasticity models | Commercial | Discover | High | Revenue | 2 | 20 | 0 | 0% |
| Engineering knowledge copilot | Knowledge | Scale | Medium | Productivity | 8 | 16 | 9 | 71% |
| Procurement document assistant | Knowledge | Pilot | Medium | Productivity | 2 | 6 | 1 | 28% |
| Driver-based forecasting | Finance | Industrialise | Low | Productivity | 5 | 9 | 8 | 88% |
| Leading-indicator safety analytics | Safety | Pilot | High | Risk | 4 | 10 | 1 | 30% |
| Alarm-flood analytics | Safety | Discover | High | Risk | 1 | 8 | 0 | 0% |
| Emissions-intensity tracker | Energy | Scale | Low | Risk | 3 | 7 | 4 | 62% |
| Carbon-intensity data service | Commercial | Discover | Medium | Revenue | 2 | 14 | 0 | 0% |
Understand. Prove. Scale.
- Business priorities
- Data landscape
- AI portfolio
- Architecture
- Governance
- Capability maturity
- Value opportunities
- AI portfolio governance
- Value measurement
- Responsible AI framework
- Data quality priorities
- Reusable architecture
- Targeted pilots launched
- Reusable AI services
- Governed data products
- AI lifecycle management
- Communities of practice
- Business-unit adoption model
Three low-risk quick wins for the first 90 days
IllustrativeChosen to build trust fast without touching operations: each is read-only or advisory, has a named owner and a baseline, and can be stopped at any time. To be validated with the business in the first weeks.
Inventory every data and AI initiative already in flight: owner, KPI, stage, risk, spend and expected value. The start of the portfolio dashboard.
- Why it is low risk
- Touches no systems; it is a governance and visibility exercise.
- Measured at day 90
- Every in-flight initiative listed with a named owner and a KPI.
The Operations Lab pattern on real historian data for a single unit, sending advisory alerts to its engineers.
- Why it is low risk
- Read-only data, advisory output, a person decides, no write path to control systems.
- Measured at day 90
- Baseline set, alerts reviewed by engineers, share of confirmed deviations.
The knowledge-assistant pattern over approved procedures and standards for one engineering team, answering only with citations.
- Why it is low risk
- Read-only, access-controlled, answers must cite approved documents.
- Measured at day 90
- Time to find information and weekly active users in the pilot group.
What I have done, and how it maps to the role.
Every claim is tied to my CV, and labelled honestly where it is transferable rather than proven.
What I have built, and why it matters here.
Former Software Developer at Securities Trading Technology (STT), Johannesburg. Software Developer 2023 – Aug 2026 · Software Engineering Intern / Scholarship Recipient 2021 – 2022. BSc Computer Science, University of the Witwatersrand.
MICS: modernised a mission-critical clearing platform from a monolith to microservices, improving scalability, fault tolerance and maintainability.
Decomposing a large system into services with clear contracts, independent deployment and owned data.
Treat AI use cases as products on a shared platform, not one-off builds.
The role, line by line.
Each requirement mapped to what I have done, what transfers and what I would do. Evidence strength is my own honest rating.
| Sasol requirement | My evidence | Transferable capability | How I would apply it | Evidence |
|---|---|---|---|---|
| Enterprise AI, Data & Digital strategy | Partnered with product owners and operations to assess options and inform architecture decisions while modernising a mission-critical clearing platform (MICS). | How a strategy becomes a target architecture and a delivery sequence without disrupting operations. | Build an Enterprise Data & AI Value Map in 90 days; link every initiative to a business KPI, an owner and an architecture standard; publish a three-horizon roadmap. | Moderate |
| Business value realisation | Translated ambiguous business requirements into working systems; owned a client-facing real-time analytics app end to end, from requirements to Google Play release. | Starting from the decision a user must make and the value it moves, not from the data or the model. | Value filter, baseline and measurement plan per use case; portfolio tracking of expected vs realised value; stop-gates for pilots that cannot prove value. | Moderate |
| Industrialise AI & data technology | Monolith-to-microservices modernisation (MICS), RabbitMQ, Docker, Kubernetes (AKS), Azure DevOps CI/CD and ARM infrastructure-as-code. | Reusable architecture, reliability and production delivery. | Establish reusable AI platform patterns — governed data products, model lifecycle, AI gateway, monitoring — with deployment standards every team uses. | Strong |
| AI & data governance | Secure REST APIs between clearing members and trading platforms (GCMS), SAML single sign-on, FIX integration, security trade-offs in a regulated environment. | Security, identity, auditability and control designed in rather than bolted on. | Responsible-AI framework, model risk tiers, model register, human-approval policy, audit logging and data classification — as shown in the Lab's governance panel. | Strong |
| Workforce enablement | Scholarship recipient who progressed from intern (2021) to software developer (2023) in market infrastructure; worked daily across product, operations and engineering teams. | First-hand understanding of what it takes for people to adopt new technology and ways of working. | AI capability ladder from aware to leader, a champions network in every business unit, playbooks, communities of practice and safe experimentation. | Developing |
| Innovation & growth | Designed and built this working application end-to-end: in-browser ML, optimisation, a grounded copilot and governance controls. | Rapid prototyping to prove an idea before scaling investment. | An innovation funnel with stage gates that looks beyond cost: data products, digital services and new business models. | Moderate |
| Centre of Excellence leadership | Led RFQ workflow delivery; engineering inside standards-driven delivery (CI/CD, IaC, shared architecture). I have not led a CoE. | Understanding, from the builder's side, why teams adopt or bypass central standards. | Run the CoE as an enablement engine: it sets standards and builds reusable capability, business teams build, governance stays consistent and value is measured. | Developing |
Three things, built on one foundation.
I understand how production systems are actually built.
I understand how technology decisions become operational outcomes.
I would apply that foundation to create reusable Data, AI and Digital capabilities.
- Engineering →
- Architecture →
- Data →
- AI →
- Digital products →
- Business value
The goal is not more AI.The goal is more value from data.
- Trusted data →
- Intelligent systems →
- Better decisions →
- Operational action →
- Measured value →
- Enterprise-scale capability
My aim would be to make Data and AI a repeatable enterprise capability — not a collection of disconnected experiments.