a lab built to
contain frontier AI.
ankya is a Melbourne research-and-deployment lab for AI, security, and cloud. Frontier intelligence is a chaotic system: powerful, nonlinear, hard to trust. We wrap deterministic rigour around it. We embed with your team, deploy models inside your own environment, and red-team them before they go live.
BOUNDED BY DEFINITION · IT EXPLORES FOREVER, NEVER ESCAPES.
MELBOURNE · −37.81, 144.96 · ABN 59 680 402 402
founder-led, security-first.
ankya is founder-led. Every engagement is run hands-on, end to end. No hand-offs, no rotating teams. That is a deliberate constraint, not a limitation. We take on one or two embedded engagements at a time, so the people who scoped your problem are the people who ship it.
The thesis is simple. Most teams bolt security and evaluation on after a model works. We treat them as the starting conditions. A deployment is not done when it produces good output. It is done when we have tried to break it and documented how it holds.
EST. MELBOURNE · AUSTRALIAN-OWNED · FOUNDER-LED DELIVERY
credentials, catalogued.
We only publish what is verifiable. No vanity metrics, no invented partner tiers. If a claim cannot be checked, it does not get a card.
Certified Information Systems Security Professional. The backbone for how we threat-model every deployment.
Qualified to audit information-security management systems against ISO/IEC 27001, the same standard we hold our own builds to.
Member of the Anthropic Cyber Verification Program, verified to conduct legitimate security research with frontier models.
PLATFORMS WE BUILD ON
four operating principles.
Research-first
Every engagement opens with analysis, not assumptions. We map the problem and its failure modes before we write code.
Precision delivery
Milestone by milestone. No scope creep, no surprise invoices. You always know what 'done' looks like.
Security-led
We red-team what we build against real threat classes: prompt injection, data leakage, model abuse, jailbreaks, insecure tool use. All of it before anything reaches production.
Knowledge transfer
Your team inherits the system and the reasoning behind it. We're done when you don't need us.
four builds, anonymised.
Client details are withheld by agreement. What is shown is the shape of the work and the real stack.
Forecasting demand for a steel manufacturer, turning historical and market signals into planning the team could act on.
Machine learning for soil and agronomy optimisation, built on a sovereign GCP pipeline.
An AI interview-preparation product: model integration, edge delivery, and the app around it, end to end.
An operations build for a moving-services company: site, tooling, and the workflow behind it.
a verifiable specimen.
Australian-owned. Founder-led. Built for teams that need AI they can actually trust in production.
ready when you are.
No pitch decks. No pressure. Just a conversation about what you are trying to build.