Applied AI Engineer
Most AI demos work once. The hard part is the system that still works on the thousandth call — under adversarial input, on real data, where a wrong answer has a cost. That layer is what I build, break, and harden.
The layer that decides if it ships
A demo proves it can work once. I build the evals, guardrails, and retries that prove it keeps working — on real inputs, at volume, when it matters.
AI agents are a fresh attack surface: prompt injection, tool abuse, data exfiltration. I red-team systems the way an attacker would, then close the gaps.
Regulated industries can't ship a black box. I make agents auditable and explainable — so a bank, insurer, or regulator can trust what they do.
We ship working AI systems you can test yourself. No testimonials from imaginary clients — just real capability you can verify.
Eight open-source AI systems, shipped with 400+ automated tests — built solo with AI-native tooling. Read every line on GitHub.
Open-source portfolio
github.com/Jbermingham1
My focus is where AI agents fail in production — prompt injection, tool abuse, data leakage. I document how systems break and how to harden them, in the open.
AI-security practice
Reliability · Security · Governance
No client logos. No invented case studies. The proof is code you can run and failures I can show you.
How I earn trust
Verify everything
If you're deploying AI agents where reliability, security, or compliance actually matter, I'd like to hear about it. A plain conversation — no pitch.
Australia-based. Focused on regulated industries.
