Work | Nicael Jooste

Portfolio

Selected work

Building digital-twin capability for water operations

Moving from dashboard thinking toward digital-twin capability for drinking-water treatment and distribution, with assumptions, validation, and ownership kept visible.

Digital-twin capability

inspectable by design

Operational data

Validation

Scenarios

Decision view

Context

Water operations contain a lot of tacit knowledge: treatment assumptions, distribution context, lab measurements, and telemetry that rarely live in one coherent workflow.

Build

I translate that context into web-based workflows for inspection, replay, scenarios, and reporting, supported by React/TypeScript interfaces, Python/FastAPI services, SDK layers, and validation patterns.

Result

The result is a shared way to look at scattered operational data — a roadmap and working patterns that domain experts, data teams, and leadership can weigh together, instead of another dashboard no one fully trusts.

Context›

Water operations contain a lot of tacit knowledge: treatment assumptions, distribution context, lab measurements, and telemetry that rarely live in one coherent workflow.

Build›

Result›

What matters next: A digital twin becomes risky if it is promoted faster than the organization can own and maintain it, so production readiness, validation, handoff, and operational ownership are part of the product.

Internal Python SDKs as data infrastructure

Reusable Python foundations that let domain experts work with enterprise data without rebuilding the same access logic in every notebook or script.

Reusable data foundation

shared interface

Data platforms

SDK

Contracts

Tools

Context

The distance between governed data platforms and day-to-day analytical work is often too large. Domain experts need stable, understandable interfaces, not just access to raw tables.

Build

I build internal SDKs above enterprise data-platform layers and APIs: standardized access methods, reusable domain logic, documentation, examples, and contribution patterns.

Result

Fewer fragile one-off scripts and faster analysis, on a shared foundation that has been adopted across a growing data team and sits cleanly between data-platform teams, data professionals, and operational users.

Context›

The distance between governed data platforms and day-to-day analytical work is often too large. Domain experts need stable, understandable interfaces, not just access to raw tables.

Build›

I build internal SDKs above enterprise data-platform layers and APIs: standardized access methods, reusable domain logic, documentation, examples, and contribution patterns.

Result›

What matters next: The long-term value is cultural as much as technical: shared interfaces make reuse, review, and documentation part of normal data work.

Governed AI adoption

Helping move generative AI from curiosity toward governed practical use: experimentation, briefings, responsible-use input, Copilot rollout support, and AI-assisted engineering.

Governed AI adoption

human review

Use cases

Guidance

Review

Practice

Context

Generative AI created immediate opportunity and immediate governance questions. In a utility context, usefulness is not enough: risk, confidentiality, and adoption patterns matter.

Build

I started experimenting with LLMs early, ran a multidisciplinary exploration group, helped shape internal use of ChatGPT and Copilot, and translated use cases for management and technical audiences.

Result

Experimentation, briefings, and responsible-use guidance built shared confidence about where generative AI helps and where it is risky — including a Copilot pilot taken from experiment to organization-wide rollout.

Context›

Generative AI created immediate opportunity and immediate governance questions. In a utility context, usefulness is not enough: risk, confidentiality, and adoption patterns matter.

Build›

I started experimenting with LLMs early, ran a multidisciplinary exploration group, helped shape internal use of ChatGPT and Copilot, and translated use cases for management and technical audiences.

Result›

What matters next: The distinction that matters is between black-box automation and inspectable augmentation: workflows where reasoning and validation stay visible.

Sector

Public references & sector work

KWR / BTO digital twin roadmap

KWR, 2023

Public sector context for digital-twin thinking in drinking-water treatment. I contributed substantial domain, product, and implementation input and am acknowledged as part of the steering committee.

KWR hydroinformatics: generative AI in water utilities

KWR, 2024

Public reference for a practical generative-AI use case in a water-utility setting, focused on extracting structure and decision support from operational plans.

TU Delft repository: field and thesis work

TU Delft

Public student-report context for earlier field and research work in water management, including mobile water monitoring in Thailand and related thesis work.

Method

How I approach the work

Utilities should buy proven systems wherever they can. The work above is what’s left when a problem is too domain-specific or exploratory to outsource — and that work only earns its place if it’s built to be owned and maintained. A few things I hold to:

Start with the operational decision, not the dashboard.
Use AI to increase throughput without hiding responsibility.
Prefer inspectable software models over impressive but opaque automation.
Treat governance, documentation, and ownership as part of the product.

Direction

What I’m interested in next

I’m interested in applied AI and data-platform work where the stakes are practical: water, infrastructure, climate adaptation, operational intelligence, and tools that help expert teams make better decisions. I especially like wrapping domain solvers such as PHREEQC and EPANET in workflows that make the inputs, assumptions, and results easy to check.

Start a conversation