Alexander
Pappas

I am a workforce analyst at Bendigo and Adelaide Bank who also ships AI products. By day I own operational analytics across six portfolios and more than 800 FTE, Adelaide first and now Melbourne. I advised executives on an operating model change that cut contact volume and staffing cost with no measurable customer impact, freeing roughly six FTE, and I replaced a proposed vendor tool with an in-house process that avoided about US$15k a year in licence cost. The recommendations I present sit behind decisions worth millions.

The rest of it is Groundwork Digital Studio, which I set up to fund my own AI usage and ended up running properly. It is where I learned to ship and then operate. Eight client products on their own domains, and underneath them an agent organisation I built that actually runs the business: scheduled pipelines into a warehouse I query, read-only canaries against each client's production database, a billing check that catches an invoice which failed to generate, and a health doctor that reconciles exit codes against assertions about the data. I built that last part after one pipeline ran green for nine days while writing nothing, because exit code zero looked like success. A job that exits clean and writes nothing now fails loudly.

What I ship is closer to product than to websites. A contract computer vision system for an external client, scoped from a single discovery call and delivered in two days as a working system plus a hosted dashboard: face detection, 128 dimension embeddings, clustering, and nearest neighbour matching against enrolled identities. It measured 0.997 V-measure, precision 1.00 and recall 0.95 on the public LFW benchmark. I label what is measured and what is simulated.

Then: a free tax tool at imsorted.au where a deterministic rules calculator checks and overrides the model's numbers before a user sees them, because a wrong number in that domain has consequences. Receipt reading and menu intake shipped into a catering client's production workflow, feeding a real business process rather than a demo. A CRM for a retaining wall business with a jobs calendar, photo capture, an SMS relay and an AI reply drafter, and a reconciler that catches an alert which silently never sent. A workforce simulator behind a versioned API, where every response carries an engine version and an input hash, and every run replays from a deterministic seed. A retrieval engine over my own archive, 4,635 chunks from about 600 files, embedded locally on purpose so a private corpus never leaves my machine. And a closed loop that scores every motion effect on this page against a measured floor before it is allowed to ship. I built that harness before I built the page.

The thread runs back further than the studio. I started experimenting with AI at the department, using it to cut the time it took to write a query and to get my head around Power BI semantic models. I was two years into a cybersecurity degree at the same time, studying alongside full time work, and that is where the foundations came from. Everything since has been the same instinct at a bigger scale: I run my own network storage and my own DNS at home, which is why the retrieval engine above embeds locally rather than posting my archive to someone else's machine.

The funding order was tokens first, then hardware. I wanted to build with these things faster than I could afford to, so I went and found work that would fund it, and the work turned into a business. What changed recently is that the tools got good enough that one person can carry a portfolio that used to need a team, which is the part I actually find interesting.

Seven years inside Australian banking and government operations sits under all of it. That is where I learned what a wrong number costs, because I was the one who had to report it. What I care about is shipping things a business actually uses. An impressive one-off is not the point. An image is an image and the gap is code, so I would rather build the thing that does the work than the thing that looks like the work.

1. Mar 2026 – nowFounder and engineer, Groundwork Digital Studio Melbourne. Eight client products built, shipped and operated solo
2. 2023 – nowSenior Workforce Analyst, Bendigo and Adelaide Bank Adelaide, then Melbourne. Six portfolios, 800+ FTE, executive and board level
3. 2019 – 2023Workforce Planning Analyst, Dept for Infrastructure and Transport (Service SA) Adelaide. Started on the counter as the centre's top performer, 30% of all customers at an eight-person site, then assistant manager, then into workforce planning across 300+ FTE at 22 metro and regional centres
4. 2017 – 2019Member Service Consultant, Credit Union SA Adelaide. Contact centre lending and insurance, fraud identification