← borck.dev
Michael Borck · Lecturer & AI Facilitator · Curtin University

Conversation, not Delegation

a way of working with AI, and the teaching, tools, and research built on it

AI should amplify human thinking, not replace it. This is a map of what I’ve built and taught at Curtin to put that into practice: small, honest, privacy-first, and useful to the people who’ll actually use it.

See the work ↓ The idea behind it
This is a map, not a brochure. Each entry leads with what it’s for and who it helps, says honestly what didn’t work. Click any card for more, then jump to the real thing.

The work

// twelve threads, one idea
Teaching & curriculum

Rebuilding what I teach, in the open

I didn't add AI as a topic. I rebuilt the teaching around it, openly, with one rule: if students can use AI they should be told how, and I should be honest about my own use too. Six units retooled, two postgraduate units and three executive courses built from scratch. Teaching the same shift to a first-year programmer and a room of executives forces a useful clarity about what's universal and what changes by level. It runs the other way too: in introductory programming, students learn to code by building AI itself, writing a chatbot and then giving it a personality, using a small package I wrote (HandsOnAI).

borck.education →
Ethics in practice

The line the AI doesn't cross

This is the spine the rest of the work hangs on, made concrete rather than just stated. The teaching runs on one rule: if students can use AI they are told how, and I am honest about my own use too. FeedForward releases feedback the academic owns, and the AI never edits student work. Signals, not scores points at evidence and never assigns the mark. Document Lens is reproducible by design, not a chatbot you have to trust. The pattern is always the same, AI assists at the boundary and the human keeps the judgment, which is the whole of Conversation, not Delegation.

conversationnotdelegation.com →
Local-AI research

Frontier-style AI on a budget, built at Curtin

It started with a constraint (no frontier cloud AI with student data) and became a research programme. An applied lab on secondhand hardware: training specialised models, benchmarking which model fits which GPU, and a free local-AI stack on the campus network. Frontier-style work on a shoestring, with a published methodology and deployed services behind it.

locolabo.org →
Tools for students

AI you can actually recommend to students

Small desktop apps that each own one job and run on the student's own laptop. Talk Buddy: rehearse a hard conversation before you have it. Study Buddy: a tutor that makes you argue for the answer instead of handing it over. Career Compass: career exploration where the résumé never leaves the machine. All free, all private.

talk-buddy →study-buddy →career-compass →all tools →
Tools for teaching staff

An AI assistant for the whole teaching lifecycle

Four tools across the teaching lifecycle, each owning one job: Curriculum Curator (design a unit), UDL Lens (audit it for accessibility), FeedForward (rubric-aligned feedback the academic releases; the AI never edits student work), and InsightLens (turn eVALUate and Insight surveys into a decision for next term). The shared posture: AI assists at the boundary; the academic keeps the judgment.

insight-lens →all teaching tools →
Assessment at scale

Signals, not scores

AI reads whatever a student submits (an essay, code, a recorded presentation, even a document's drafting trajectory) and maps what it finds to your rubric as observations, not grades. You read the signals, weigh them, assign the mark, and write the feedback. It makes a cohort of 300 readable without the AI ever marking, and assessing the evidence and the process is more defensible in an AI world than marking a final text. A separate strand asked the question directly, can AI grade at all? Benchmarking it against human markers (AssessmentBench) showed it is good at the average and bad at the precise, which is exactly why the answer is signals, not scores.

github.com/michael-borck →
Governance

AI in teaching without sending data anywhere

Most institutional AI adoption stalls on one question: where does the data go? The answer here is 'your own machine.' The case (and the working tools) for small, single-purpose, privacy-first AI that runs locally, with no telemetry and no shared cloud database. As subscription and token costs climb, local-first is increasingly the sensible default for education.

locolabo.org →
The role

A year as a school’s first AI facilitator

A candid account of a year as a school's first embedded AI Facilitator: how staff actually used the time (mostly not what you'd predict), what the Skills Passport and the AI Exchange (a verified-anonymity space for staff to share what's actually working with AI) changed, where workshops landed and where they flopped, and the honest finding: a single embedded person doesn't scale, and some of what worked, worked because it was one trusted person in the corridor.

borck.education →
Building method

What AI lets one academic build, and where it breaks

About thirty tools shipped solo, in conversation with AI, and an honest account of the failure modes: confidently-wrong code, the pull to delegate instead of think, the maintenance trap. The breadth is evidence, not the point; the working method and the disciplines that keep it sustainable are. The raw portfolio lives on GitHub; the polished subset is on the tools site. One example of the range: a small media pipeline that turns a markdown deck into a narrated video with sourced images, an avatar walk-on, and a chatbot grounded in the unit's documents, no recording studio required.

github.com/michael-borck →tools.borck.education →
Authoring

Books written the way I tell students to

Two framework books and a verb-by-verb Python series: published, open-licensed, and usable in your own units. An honest look at writing them with AI: what it was genuinely good at, where the human had to hold the line on voice and the examples that matter, and the publishing reality nobody mentions.

books.borck.education →
Simulations

Rehearse the situation before the situation

We send students into a placement, client project, or capstone having never rehearsed it, and that's hard to arrange and doesn't scale. So build the rehearsal. A platform that generates realistic workplace simulations (a company, AI 'employees', internal documents, real tasks) so a whole cohort can practise the role and the workflow before the real thing. A simulation isn't the real thing, but practice is practice, and rehearsed students show up ready. One such world, CloudCore (a firm recovering from a data breach), is taught across five units, with AI staff who only answer during business hours and a story that time-releases as the term unfolds.

locoensayo.org →simulations.borck.education →
Open by default

Built to give away

Almost everything here is free, MIT-licensed, and public on GitHub: the tools, the teaching package, the simulation platform, the books. The point isn't generosity for its own sake. Open and local-first is what clears the institutional bar, with no procurement, no shared cloud database, and no data leaving the machine. It also keeps me honest: if a colleague can read the code and run it themselves, the work has to actually hold up. The polished subset lives on the tools site; the raw portfolio is on GitHub.

github.com/michael-borck →tools.borck.education →

Built with colleagues · used across Curtin

The best of this work isn’t mine alone: it’s co-taught, adopted, and fronted by the people who use it. That’s the point.

What I’m exploring now