Every business is going to end up running its own intelligence layer. Not a chatbot bolted onto a website - something that knows how that particular business actually works, sits on its real data, respects who is allowed to see what, and can be held to account when it gets something wrong.

Most businesses are nowhere near that yet. The gap is almost never the model.

Tessellant builds that layer for small and mid-sized Australian businesses. We are small on purpose, and we are booked. What we want now is a bench: a short list of people we trust and can call when the work needs more hands.

To be clear about where this sits. We are not filling a seat this month. We are opening a two-week window to meet good people, understand what you want the next few years to look like, and start the relationship before we need it. If that turns into work, good. If it turns into one useful conversation and nothing else, that is a fair outcome too.

How we think about AI

- Deterministic where it matters. Anything running unattended has to be predictable, testable and auditable. Models are strong in the fuzzy middle and untrustworthy at the edges. We design around that rather than hoping.
- The unglamorous work is the work. Access, identity, data quality, permissions, the integration nobody wants to own. Projects stall there, not on model choice.
- Context, not data. Businesses have plenty of data. What they lack is the context that makes it mean anything.
- Small, provable, then wider. One team, one process, one week. A live thing somebody actually uses beats a slide about what is possible.
- Vendor-neutral and portable. We are not resellers. Clients own their layer, their data, and their exit.

Who we want on the bench Three profiles.



You do not need all three - most valuable people are deep in one and dangerous enough in another.

AI and platform engineers

- using technologies such as Python (FastAPI, Pydantic), Postgres including pgvector and row-level security, containers, CI/CD. Model gateways and routing, MCP servers, evals and tracing. Identity done properly: Entra ID, OIDC, OAuth2. You have shipped something a real person used, and you can explain why it broke the first time.

Data and integration - i ngestion pipelines (dlt or equivalent), warehouse modelling, semantic layers, REST and OData integration against practice-management, finance and line-of-business systems.

Microsoft 365 and Graph, Power Platform. You are comfortable when the source system is badly documented, and you test what the vendor tells you rather than taking it on faith. Delivery and analysis - s

itting with a client, watching how the work genuinely happens, and coming back with what is worth building. Scoping, requirements, and writing that a business owner will actually read. Enough technical depth to know what is cheap and what is not.

What working with us looks like

Remote-first across Australia. Small team, short cycles, real deployments into real environments. You would own a piece of work end to end rather than sit in somebody else's queue.

How to apply

Send your CV with a short note. A paragraph is plenty. Cover three things:

1. Something you have built that you are proud of, and what you would do differently now
2. What you want the next two or three years to look like
3. Where you are based, and what your availability looks like

We will read everything that comes in over the fortnight and book calls with the people we want to know better. Those calls are conversations, not interviews.

📌 Building a bench - AI engineering, data and delivery (Australia)
🏢 Tessellant
📍 Australia