Sources

Pattern of early adoption: Vue 2015. Remote 2017. Local AI 2022. MCP 2025. Governance 2025. Vue.js in 2015, before it was the obvious choice. Remote work in 2017, before the world caught up in 2020. Local AI models in 2022, before consumer hardware made it routine. MCP servers in early 2025. AI governance also 2025. The pattern is timestamped. Ideas tend to predate market validation here by one to three years.

Client timeline: Chronology of the engagements that landed. ACER (2014). Coles (2014). ESPN (2015). Tabcorp (2016). Esri (2017). Buddy Platform (2018). Prime Motive (2018). RMIT Online (2018 to 2019). NAB (2019). Morgan Stanley (2019 to 2020). Medibank (2020). World Vision (2020 to 2021). Pearson (2021). Vanguard (2021 to 2023). ADF Careers (2023). x-RD (2024). Australian Defence Force (2024 to 2025). Mentorloop (2025).

Contact: robert@textual.au robert@textual.au Melbourne, also Sydney and Canberra. Available for contract, permanent, or advisory. Defence and government work history available on request.

Cross-domain transfer: One pipeline pattern, four domains. A signal-score-rank-output pipeline shows up across multiple unrelated builds: financial fraud detection, recruiter candidate scoring, candidate tracking, preference extraction. Same architectural shape, different inputs and weights. The pattern is not a framework. It is the way a certain class of problem decomposes when you read it correctly.

Disciplines: Design and engineering, both from 2004. Design and engineering, both from 2004. The thread between them isn't versatility, it's that the same patterns travel between substrates. Design system work has carried through RMIT, Vanguard, Pearson, Medibank, and ADF Careers. Engineering through everything else.

Early career: From clan sites at 12 to Textual in 2014. First websites at age 12, Counter-Strike clan sites. Everex Media founded at 14 in 2004, run through to 2009. Swinburne University for Business and Design, deferred, not completed. Textual consultancy from 2014.

Advisory engagement: Principal on call, minimum four days a month. Advisory is principal-level consultation: architecture review, code review, technical direction, hiring support. The team builds; the principal reviews and steers. Minimum four days per month. No advisory projects on record yet.

Embedded engagement: Day-rate, inside the client team, minimum two weeks. Embedded means working inside the client team's tools, processes, and cadence. Slack channels, sprint ceremonies, shared accountability. Day rate, minimum two weeks. Most of the historical work has been embedded because clients generally want senior capability inside the team.

Scoped engagement: Fixed scope, fixed price, handover on completion. Scoped means a defined piece of work, a fixed price, and a clean handover at the end. Useful when the brief is tight and the team is in place. The work gets defined first, then built, then handed over. No recent scoped projects on record.

Engineering evaluation: Five diagnostic lenses for deciding whether code earns its place. Five diagnostic lenses, applied in this order: 1. Deletion test: who notices if this is removed? 2. Replacement test: could something simpler do 90% of the job? 3. New hire test: understood in under an hour? 4. Scale test: justified by current scale? 5. Abstraction test: how many concrete implementations does this serve? No-stopgap rule. If it cannot be built properly now, leave it unbuilt with a comment. A stopgap that ships becomes the system.

Industry experience: Defence, finance, healthcare, education. Defence, with Baseline clearance. Finance, on institutional interfaces. Healthcare, on national-scale component systems. Education, on learning analytics and accreditation platforms. Each one has its own constraint shape. Defence is clearance and isolation. Finance is regulated UI and audit trails. Healthcare is multi-tenant content systems. Education is competency tracking that has to outlast individual courses.

Methodology: Define business requirements and constraints before selecting technology. Approach: define business requirements and constraints before selecting technology. Four-stage sequence: separate intent from stated ask, understand the ask, reconcile at minimum scope, iterate the build. Practices: config-driven rules, phased execution with verification gates, domain-transferable architecture. Scientific method applied to product decisions, not as metaphor: hypothesis, null hypothesis, test, measurement.

Tech stack: What I reach for, by layer. **Frontend.** Vue, Nuxt, React, Next.js, React Native and Expo, Angular, Stencil. **Backend.** TypeScript, Node, Express, Hono, Python, Django, FastAPI, Bun, Rust. **AI and ML.** AWS Bedrock, LangChain, LangGraph, PyTorch, MLX, the Anthropic API directly, vector embeddings and RAG. **Data.** PostgreSQL with pgvector, Elasticsearch, SQLite with FTS5, MongoDB. **Infra.** Docker, Azure DevOps, GitHub Actions, Vercel, Fly.io, Tauri, RunPod.

ADF Careers · Australian Defence Force · Senior Frontend Engineer · Aug 2023 – Nov 2023. Engagement: embedded. Frontend and design-system work for the ADF recruitment hub.. Tech: React, Next.js, Redux, TypeScript, Storybook.

ADF Careers · Australian Defence Force · Senior Frontend Engineer · Aug 2023 – Nov 2023, lede: Three months on the ADF recruitment hub. React and Next.js on the front, Redux behind it. Component patterns, Storybook documentation, and consistent interface conventions across services. Recruitment for the Australian Defence Force runs across services, role families, and eligibility paths that don't collapse into a clean wizard. The interface had to absorb that without making the candidate feel it. Embedded engagement, working alongside the in-house team. The brief was clearer interfaces and consistent patterns. What shipped was a tighter contract between the design intent and the front end, so the next person opening the codebase didn't have to guess which button was canonical. Separate engagement from the later Defence training platform. Different product, different team, different year. Same client.

Australian Defence Force · Principal Engineer · Dec 2024 – Jul 2025. Engagement: embedded. Competency platform at the centre of a Defence training initiative, built on public standards. Search and learning analytics across the framework.. Tech: Vue 3, Nuxt, Elasticsearch, Keycloak, Azure AD, xAPI, CASS (IMS Global).

Australian Defence Force · Principal Engineer · Dec 2024 – Jul 2025, lede: Seven months as principal engineer on an ADF training platform. Baseline cleared. The brief was tracking demonstration, not completion. That distinction shaped almost every other decision.

Australian Defence Force · Principal Engineer · Dec 2024 – Jul 2025, standards over custom: The data layer speaks public standards from IMS Global: CASS for competency modelling, xAPI for the analytics. Picking standards meant the platform spoke a public language from day one rather than a bespoke vocabulary that would need a translation layer at every integration point. The internals of how the standards are applied stay private.

Australian Defence Force · Principal Engineer · Dec 2024 – Jul 2025, what it does: Find a person by competency, or a competency by person. Elasticsearch under search, the graph layer in front. Relevance had to absorb the framework's depth, where what mattered for someone's progression wasn't always what matched the keyword. xAPI gives you the demonstration, not just completion. For Defence, completion was never the question. The question was whether the person could do the thing, and whether you could prove it later to someone who wasn't in the room.

Australian Defence Force · Principal Engineer · Dec 2024 – Jul 2025, identity: Keycloak federated to Azure AD so personnel signed in with the credentials they already had. The configuration is the part I can't talk about. Standard Keycloak shape on the public side; the contract takes over from there.

Australian Defence Force · Principal Engineer · Dec 2024 – Jul 2025, outcome: Owned the build end to end. Real-database integration tests from day one. Stayed inside the lines of what adjacent teams owned. Seven months. Handed over running.

Project Eucalyptus · Aliases: Euco, Project Euco, Eucalyptus, Project Eucalyptus · Textual · Researcher · 2025 – ongoing. Engagement: research. Eucalyptus waste to multiple higher-value product lines. Pre-publication research..

Project Eucalyptus · Aliases: Euco, Project Euco, Eucalyptus, Project Eucalyptus · Textual · Researcher · 2025 – ongoing, lede: Eucalyptus waste, run through pyrolysis, can be coaxed into several distinct higher-value product lines. The interest is in turning one overlooked feedstock into a portfolio of viable businesses. Pre-publication.

Project Eucalyptus · Aliases: Euco, Project Euco, Eucalyptus, Project Eucalyptus · Textual · Researcher · 2025 – ongoing, the setup: Australia has plenty of eucalyptus. Plantation residue, sawmill offcuts, post-fire salvage, prunings that currently go to chip or compost or burn pile. Pyrolysis itself is well-known: heat biomass without oxygen, get char, oil, and gas back. The chemistry has been understood for a century. The work that matters is elsewhere, in the product mix that comes out the other end and what you do with each fraction. Most facilities in this space are tuned for a single output. Optimise for one product, the rest becomes byproduct, and byproduct is the part of the spreadsheet that quietly kills the economics. The framing here inverts that: portfolio over single-product. Tune the run for the right mix and each fraction becomes a separate revenue line. Several lines under the same feedstock, sitting across distinct end markets.

Project Eucalyptus · Aliases: Euco, Project Euco, Eucalyptus, Project Eucalyptus · Textual · Researcher · 2025 – ongoing, why now: Two reasons. The first is feedstock. Australia sits on enormous volumes of eucalyptus residue that already exist as a logistics problem before anyone has tried to make anything from them. Plantation thinnings, sawmill offcuts, post-fire salvage, the prunings that local councils pay to remove. The supply curve is steep on the wrong side, so the input cost of the raw material can be negative once you account for what people are paying to make it disappear. The second is sovereignty. A non-trivial fraction of what Australia imports could plausibly be produced from this feedstock domestically. That's a research conclusion. The policy conversation around it is the reason ASPI is in the loop. The import dependency itself is what gives the project its civic weight.

Project Eucalyptus · Aliases: Euco, Project Euco, Eucalyptus, Project Eucalyptus · Textual · Researcher · 2025 – ongoing, what i can say: The framing is portfolio over single-product. A facility tuned for one output discards the rest. A facility tuned for the right mix turns each fraction into a separate revenue line, with the byproduct economics flipped from drag to lift. The numbers behave very differently once that reframing holds. The work to date has been desk research, pyrolysis literature, feedstock surveys, end-market sizing, and the long unglamorous task of working out which lines hold up under realistic input prices and which were always going to be a story. Most of the engineering questions are unsexy. Particle size and moisture in the feedstock. Residence time and temperature profile in the reactor. The condensation train on the oil side. Catalyst choice if catalyst is on the table. None of it is novel chemistry. The novelty is in how the lines compose and what each one is worth at the gate.

Project Eucalyptus · Aliases: Euco, Project Euco, Eucalyptus, Project Eucalyptus · Textual · Researcher · 2025 – ongoing, what i can't say: The product lines themselves. The patent claims. The process specifics. The commercial strategy and the business plan. All pre-publication, all sealed until publication. ASPI outreach is part of the pathway. Beyond that, the conversation goes offline. The redactions here are doing real work. The gap between "interesting research" and "filed claim" is the part of the project that has to stay on paper, in counsel's hands, until the right moment.

Project Eucalyptus · Aliases: Euco, Project Euco, Eucalyptus, Project Eucalyptus · Textual · Researcher · 2025 – ongoing, why this is on the portfolio: Most engineering portfolios don't include chemistry. This one does because the patterns are the same: read a system carefully, find the part that's been miscategorised, run the numbers, name the opportunity. The substrate changes. The thinking does not. Software, identity, agentic AI, biomass. Same instinct. The byproduct is the business.

Hawk · Textual · Sole developer · Jan 2025 – Jun 2025. Engagement: product. Multi-tenant AI coding middleware. OpenAI-compatible endpoint for IDEs. Retired.. Tech: Rust, Node, PostgreSQL, React, LangGraph.

Hawk · Textual · Sole developer · Jan 2025 – Jun 2025, lede: A multi-tenant AI coding middleware that learned how developers work, routed requests across providers, and exposed an OpenAI-compatible endpoint for IDEs. Profile learning, context filtering, and budget-aware orchestration sat underneath it. Retired. Implementation details, orchestration architecture, and the profile-learning mechanics are off-limits.

Kernel · Textual · Researcher · 2025 – ongoing. Engagement: research. OS-level resource management research. Quantum state models. Pre-publication..

Kernel · Textual · Researcher · 2025 – ongoing, lede: Quantum state models inside OS-level resource management and process isolation. The interest is in rethinking how identity and state shape what processes can see, reach, and share. Still early. Research specifics are sealed until publication.

Medibank · Senior Software Engineer · May 2020 – Sep 2020. Engagement: embedded. Component toolkit for Adobe Experience Manager. React-backed AEM components inside the CMS.. Tech: React, AEM, AEM Components.

Medibank · Senior Software Engineer · May 2020 – Sep 2020, lede: Four months on a component toolkit for Adobe Experience Manager. React-backed AEM components, designed to behave cleanly inside an enterprise CMS instead of fighting it on every page. The point was reuse. AEM is not a forgiving environment. Authors edit content in one place, components live in another, and the boundary between the two is where most enterprise component libraries quietly fall apart. The work was making React-backed AEM components that survived authoring, survived the CMS lifecycle, and stayed predictable for a content team shipping every week. Embedded engagement. Senior Software Engineer inside the existing team. Built a set of building blocks the rest of the organisation could trust, with the AEM-specific edges sanded off so the developers consuming them didn't have to learn the platform first.

Mentorloop · Senior Product Engineer · May 2025 – Dec 2025. Engagement: embedded. Embedded product engineering on Mentorloop's mentoring platform, plus Loopy, in-app AI chat on AWS Bedrock that adapts to context.. Tech: AWS Bedrock, LangChain, Meteor, React, MongoDB, Lexical.

Mentorloop · Senior Product Engineer · May 2025 – Dec 2025, lede: Eight months embedded at Mentorloop. Contracted for general product and UI feature work on the mentoring platform. The AI piece, Loopy, came out of that engagement rather than arriving as a separate brief.

Mentorloop · Senior Product Engineer · May 2025 – Dec 2025, the setup: Mentorloop runs a global mentoring platform. The product wraps mentoring programs end to end. I joined the team for general product and UI work; Loopy followed.

Mentorloop · Senior Product Engineer · May 2025 – Dec 2025, the product work: Most of the eight months sat inside the team's existing cadence: feature work across the platform, accessibility improvements, internationalisation, identity integrations, and editor work. The work fit into the existing release process rather than running parallel. Implementation specifics stay internal.

Mentorloop · Senior Product Engineer · May 2025 – Dec 2025, loopy: In-app AI chat for coordinators running mentoring programs. Route-aware and role-aware, so the same chatbot answers participants and coordinators in the right vocabulary for what they are doing in the product. Built on AWS Bedrock. Multi-lingual from day one. Operational tuning lived in client tooling so the team could adjust behaviour without a deploy. Other AI features at Mentorloop were built by separate teams; the AI piece I built was Loopy.

Morgan Stanley · Lead Design Engineer · Nov 2019 – Mar 2020. Engagement: embedded. Frontend architecture and visual system for an institutional finance concept.. Tech: Angular, TypeScript, Sass.

Morgan Stanley · Lead Design Engineer · Nov 2019 – Mar 2020, lede: Five months helping establish the frontend architecture and visual system for an institutional finance concept. Angular, TypeScript, Sass. Regulated environment. Clarity carried the system. The work was about making the structure of the application visible at a glance, so traders and operations staff could see the state of the world without having to interpret it. Most of the design decisions were subtractive. Internal Morgan Stanley processes are off-limits. The public version is what the system looked like and how it was built, not how it was run.

National Australia Bank · Lead UI Designer · Jun 2019 – Nov 2019. Engagement: embedded. UI design for a financial reporting tool, plus a small internal Vue app for engineering.. Tech: Figma, React, Vue, Compliance Tooling.

National Australia Bank · Lead UI Designer · Jun 2019 – Nov 2019, lede: Five months at NAB. UI design for a financial reporting tool, plus a small internal Vue app for the bank's engineering teams. Figma on the design side. React and Vue downstream depending on which team was consuming the work. The reporting tool was the larger piece: design that had to survive a compliance review, then translate cleanly into a product that finance staff actually wanted to open. The internal Vue app was smaller and faster, an engineering-facing piece where the audience could read the screen the moment it loaded. Embedded engagement. Design that respected the technical reality of the teams shipping it.

Pearson Education · Product Design Lead · May 2021 – Nov 2021. Engagement: embedded. Design systems and implementation-ready specs for education products. WCAG 2.1 from the start.. Tech: Figma, Design Systems, Interaction Design, WCAG 2.1.

Pearson Education · Product Design Lead · May 2021 – Nov 2021, lede: Six months on design systems and implementation-ready specs for education products. Accessibility was part of the system from day one, WCAG 2.1 baked into the components themselves, so consuming teams couldn't ship inaccessible by accident. Assessment patterns were the other thread: how you surface right/wrong, how you show progress to a student who needs to understand where they stand. The components carried those decisions so the teams consuming them didn't have to remake them on every screen. Interaction design lived in the same place as the specs. Embedded engagement, product design lead.

Prime Motive · Prime Motive / CGU Insurance · Senior Front End Engineer · Oct 2018 – Dec 2018. Engagement: embedded. Campaign site for CGU Insurance on a tight timeline.. Tech: Nuxt, Vue, JavaScript, Sass.

Prime Motive · Prime Motive / CGU Insurance · Senior Front End Engineer · Oct 2018 – Dec 2018, lede: Three months on a campaign site for CGU Insurance. Tight timeline. The brief was high-motion presentation that stayed performant on the phones the campaign was actually going to land on, which were not flagships. Nuxt and Sass kept the bundle honest. Most of the work was deciding what to cut so the rest could feel expensive on a four-year-old Android. Motion that costs nothing to render reads better than motion that costs everything and stutters. Senior Front End Engineer, embedded inside the agency team. A campaign site that didn't fight its own audience.

RMIT Online · Senior Designer / Lead Frontend Engineer · Dec 2018 – Jun 2019. Engagement: embedded. Design and frontend rebuild for the RMIT Online website. Vue, Nuxt, GraphQL.. Tech: Vue, Nuxt, GraphQL, WCAG 2.1, Vercel.

RMIT Online · Senior Designer / Lead Frontend Engineer · Dec 2018 – Jun 2019, lede: Six months at RMIT Online, on both sides of the line. Design first, then lead frontend through a full rebuild in Vue and Nuxt with GraphQL underneath. Multi-cohort education platform. Two roles, one codebase, one product. The bridge between design intent and shipped code stayed short because the same person was on both ends of it. Search and filtering were the centre of it. An education platform serves students across different programs, enrolment states, and course types. The query surface had to absorb that without making the student feel the complexity underneath. WCAG 2.1 was part of the rebuild from day one, treated as architecture. Vercel on the deploy side. Embedded engagement.

Vanguard · Senior Engineer / Product Architect · Dec 2021 – Jun 2023. Engagement: embedded. Two separate engagements across eighteen months. Stencil web components for multiple consuming teams, then a React Native library for iOS and Android.. Tech: Stencil, React Native, TypeScript, Design Systems.

Vanguard · Senior Engineer / Product Architect · Dec 2021 – Jun 2023, lede: Eighteen months across two Vanguard engagements. First web components. Then mobile components. Different teams, different stacks, same pressure: make the interface language hold together across product surfaces.

Vanguard · Senior Engineer / Product Architect · Dec 2021 – Jun 2023, engagement one: stencil: A web component library with several frontend teams consuming it. The internal stack already ran React, Angular, and vanilla JS, so the library needed to meet each team where it was. Stencil gave the system a framework-neutral base: build the component once, consume it from whichever application needed it. Adoption was the work. Each team had its own theming, release cadence, and composition habits. Storybook held the contract: specs, states, usage notes, and examples in one place, with the shared component kept small enough for consuming teams to trust.

Vanguard · Senior Engineer / Product Architect · Dec 2021 – Jun 2023, engagement two: react native: A React Native component library backing both the iOS and Android apps from a single codebase. Different consumers, different design tokens, different release process. The mobile work kept parity visible in the code before it could drift in the product. Buttons, sheets, typography, spacing, loading states. Two platforms solving the same problem twice was the pressure. The component library carried that discipline. Tokens at the bottom, primitives above, composed components above that. Platform-specific code was allowed where it made sense, but it lived inside the component rather than leaking into every consuming screen.

Vanguard · Senior Engineer / Product Architect · Dec 2021 – Jun 2023, tokens, primitives, composition: The architecture stayed deliberately plain. Design tokens generated from the source of truth, primitives consumed the tokens, composed components consumed the primitives. Three layers, narrow contracts, predictable change. The boundary between layer two and layer three was the call to make. Primitives stayed framework-neutral: a Pressable, a Text, a View with the right defaults. Composed components carried product meaning: a transaction row, a chart card, a confirmation sheet. When a component started carrying too much behaviour, it got split before that behaviour hardened into API. That kept the library maintainable. New screens always ask for variants. The useful answer was usually composition, not another prop.

Vanguard · Senior Engineer / Product Architect · Dec 2021 – Jun 2023, what lasts: Led three to five engineers across the two engagements. Eighteen months total. Practical system work: small contracts, documented states, and components that consuming teams could adopt without ceremony. Internal Vanguard process, team names, individual names: not on a public portfolio. The craft side is here. The rest stays internal.

World Vision · Lead Design Engineer · Oct 2020 – Apr 2021. Engagement: embedded. White-label donation platform for partner organisations running their own branded versions on top of World Vision's infrastructure.. Tech: Nuxt, Vue, Braintree, Multi-tenancy.

World Vision · Lead Design Engineer · Oct 2020 – Apr 2021, lede: Six months building a donation engine that had to look like several different partner organisations from one codebase. Nuxt on the front, Braintree on payments, multi-tenancy in the data layer rather than scattered through controller logic.

World Vision · Lead Design Engineer · Oct 2020 – Apr 2021, the tenant model: The tenants weren't World Vision regions or country offices. They were partner organisations, larger corporate or charity donors that ran their own branded donation experience on top of World Vision's infrastructure. Each got its own surface, its own brand, its own donor relationship. World Vision provided the engine. Tenant configuration sat in data. Each partner got its own row, its own brand identity, its own copy, its own payment processor settings, its own receipt templates. Onboarding a new partner was a config change, not an engineering project. That distinction is the whole story. When the configuration is the contract and the code is the engine, the team running partner relationships can spin up a new branded surface without filing a ticket, and the engineering team isn't pulled into every brand refresh. The trap in white-label work is letting tenants leak into the codebase. One conditional becomes ten. Ten becomes a graveyard of `if (tenant === 'xyz')` blocks that nobody understands six months later. The discipline was to refuse that drift on day one. If something needed to vary by partner, it varied through configuration. If it couldn't vary through configuration, the question was whether the variation was worth keeping.

World Vision · Lead Design Engineer · Oct 2020 – Apr 2021, rendering by partner: Each tenant rendered its own surface from a shared component layer. The components didn't know which partner they were inside. The tenant context did the lifting: theme tokens, copy strings, payment methods, receipt formatting. Nuxt handled the routing and the SSR layer cleanly. The donor on the other end saw the partner's brand, the partner's framing of the cause, the partner's payment options. Underneath, a single render path. Each partner organisation had its own identity and its own editorial voice. The platform respected that. A donor giving through one partner was supposed to feel like they were on that partner's site, not on a generic donation form with someone else's logo bolted on. The visible surface had to feel like the partner's. The infrastructure underneath had to be unified. Both things at the same time, without compromise to either.

World Vision · Lead Design Engineer · Oct 2020 – Apr 2021, braintree, plumbed properly: Payments went through Braintree. The work there was less about integration and more about isolation. Each partner had its own merchant account, its own fraud rules, its own receipt templates. The donation flow normalised the inputs and routed to the right account based on tenant context. Failure modes were explicit: card declined, processor timeout, fraud flag. Each one had a path back to the donor that respected the moment they were in. People donating in response to a crisis don't need a generic error page. Recurring donations carried their own state machine. Sign-ups, renewals, failed retries, cancellations, lapse recovery. The tenant layer extended through that state machine too, because retention copy and retry cadence varied by partner.

World Vision · Lead Design Engineer · Oct 2020 – Apr 2021, what i can't say: Donation volumes. Fundraising numbers. The dollars are off-limits.

World Vision · Lead Design Engineer · Oct 2020 – Apr 2021, outcome: Shipped. Multiple partner organisations running through one codebase, each one looking and behaving like its own thing. The configuration-driven tenant model held up through new partner onboarding without code changes, which was the bet at the start of the project and the proof at the end of it.

x-RD · Senior Full Stack Engineer · Jan 2024 – Sep 2024. Engagement: embedded. Defence logistics platform. Nine months embedded. Frontend and process layers.. Tech: React, Next.js, Python, Django, Zebra.

x-RD · Senior Full Stack Engineer · Jan 2024 – Sep 2024, lede: Nine months on a Defence logistics platform. Baseline cleared. Front-end and process-layer work, with Zebra integration inside the labelling flow. The redactions on this engagement are wide enough that the public version is almost the whole story. Defence logistics, React and Next.js on the front, Django on the process side, Zebra in the warehouse loop. What I can't say covers the rest of it. Architecture, deployment, hardware, network, team, constraints, scope, outcomes, client specifics, problem domain. That conversation is offline.