Melbourne, Australia
Principal-led AI delivery, platform engineering, and data strategy for Australian organisations. We build systems your team can run without us, and we document them so they stay that way.
Most problems arrive as a mix of all three. We scope across them rather than selling you the one we happen to staff.
Take AI past the proof of concept. Retrieval systems grounded in your own data, models that hold up in production, and an honest view of what the technology will and will not do for your use case.
Move legacy estates onto cloud-native platforms without stalling the reporting the business already depends on. Migration, pipeline architecture, and the automation that removes manual handling.
Translate business goals into a data capability plan that survives contact with a budget cycle. Prioritised initiatives, a target architecture, and the governance to keep it trustworthy as it scales.
Some problems need a defined scope and a fixed outcome. Others need senior capacity inside your team for a while. We do both, and we will tell you which one your problem actually is.
A defined scope with an agreed outcome, a fixed price or capped time and materials, and a handover at the end. Best when the destination is clear enough to write down and you want someone accountable for reaching it.
Senior data and AI capability inside your existing team, on a day rate, for a defined period. Best when you have the direction and the backlog but not the people, or when the work needs specialist depth alongside your engineers rather than instead of them.
Larger scopes are staffed by bringing in specialists against a defined plan, with the principal accountable for delivery end to end.
Engagements are described by sector and scale rather than by client name. Happy to walk through any of these in detail on a call.
MVP delivered ahead of schedule
Decades of business logic sat inside COBOL and SAS code that predated most of the people supporting it. The rules governing claims and pricing existed nowhere else, so any move to a modern platform risked silently changing how the business worked.
Architected a platform that uses LLMs to reverse-engineer the business rules out of the legacy code, extract data lineage across the estate, and generate the transformations onto Databricks and Snowflake. Built as a production system with human review at each step, not a one-off extraction script.
MVP delivered a month ahead of schedule, with previously undocumented rules captured in a form the business could read and challenge. Migration could then proceed against a known specification rather than an assumption.
Manual reporting effort removed
Reporting across more than 300 venues was assembled by hand every week. Executives waited days for numbers, and by the time a trend was visible the window to act on it had usually closed.
Automated the reporting layer end to end, then led the business side of migrating the legacy estate onto Databricks, building shared and reusable data assets rather than another set of per-team extracts.
Manual workload down by roughly 70 percent, over 20 hours a week returned to the team, reporting turnaround halved, and query performance around three times faster on the migrated platform.
No document or query leaves the environment
The answers people needed were spread across a large internal document estate, and keyword search returned documents rather than answers. The material was sensitive enough that sending it to a commercial model API was never an option, which ruled out most of the tooling on the market.
Built the retrieval platform end to end: ingestion, hybrid semantic and keyword search, and cross-encoder reranking before anything reached the model. Answers are generated by open-source models running inside the client's own environment, deployed on Azure behind a lightweight front end.
Staff can ask questions in plain language and get answers traceable back to the source document, with nothing leaving the tenant. The privacy constraint that would have blocked a hosted API became a design input rather than a reason not to build.
Manual compliance review effort
Compliance review depended on people reading large volumes of documents and correspondence by hand. Coverage was capped by how much a team could physically get through, so breaches surfaced late or not at all.
Built and deployed an NLP system that classified the material and flagged it for review, prioritising the cases most likely to contain a breach instead of sampling at random.
Manual review workload down 25 percent and materially better breach detection, with reviewers spending their time on the cases that warranted it.
Upskilled, with a privacy position in place
Generative AI arrived faster than the policy around it. Teams were keen to use it on client work, but nobody could say what was safe to put into a model or which use cases would stand up to scrutiny.
Ran organisation-wide research into the tooling, completed the privacy assessments, and turned the findings into practical training rather than a policy document nobody would read.
More than 60 staff upskilled, a clear internal position on what was permitted, and capability aligned to the work clients were beginning to ask for.
Four stages. The last one is the point: if your team cannot run it after we leave, the engagement did not work.
Stakeholder sessions and a technical assessment of the current estate. We are looking for the constraint that actually binds, which is rarely the one in the brief.
A prioritised roadmap and target architecture, with the trade-offs written down. You get a clear view of what we would not do, and why.
Delivery in short cycles with working software at the end of each one. Your team sits in the reviews from the first sprint, not the last.
Production deployment, runbooks, and structured knowledge transfer. We leave behind something your team owns rather than something they call us about.
The practice is principal-led. The person on the first call stays on the engagement, in the architecture and in the code.
OZ Data Solutions was founded after a decade delivering data and AI inside Big 4 consulting and ASX-listed enterprise, with experience gained at Deloitte, EY and Tabcorp. The principal holds a PhD in computer science from Monash University and stays hands-on: architecture, code, and the executive conversation that decides what actually gets built.
He is currently engaged with a major Australian health insurer, leading an AI portfolio that spans an LLM-based modernisation platform reverse-engineering business rules out of COBOL and SAS estates, a suite of insurance applications covering fraud detection and claims operations, and a retrieval platform running entirely on local open-source models for privacy.
Every claim on this page is checkable. These are the facts behind them.
Whether the brief is written or still a vague worry, a first call costs nothing and usually clarifies more than a proposal would.