AI Learning Architect
Instructional Design + Context Engineering for Learning Systems · Sydney
I design and build learning systems that run on AI. That covers agent architecture, Claude API orchestration and the contract and retrieval work that decides what a system should do before anyone builds it. It's the pipeline and the platform. Not just the content that runs on them.
That work sits on 15+ years across enterprise insurance, rail safety, financial services certification and bilingual digital learning for NGOs and government. These are domains where you can't fake compliance and you can't fake the numbers. AI lets me move faster on content production without losing the governance those clients need.
I live in Sydney and work remotely. I'm a permanent Australian resident and I deliver in English and Spanish for LatAm-facing programs and CALD workforces.
I work out what the knowledge layer should do before anyone builds it. Retrieval contracts, primitive selection, output shapes, provenance. I translate SME expertise into something the agent can actually act on.
Schema-driven curriculum ops, regulatory framework translation, outcome tracking and evidence systems. Mapping content to formal standards is knowledge-graph design with a different name.
Multi-call Claude pipelines with human-in-the-loop gates. Full-stack platforms on Next.js, Supabase, Stripe and Vercel. Airtable and Softr for lighter builds.
QA gates, validation and provenance work, handover docs. I apply the regulated-industry reflex to AI systems. That means protecting an agent's context from rot, not just shipping a working demo.
A learning operations platform that maps evidence to NESA Stage 4 outcomes. 179 outcomes across seven subjects, dual Standard and Life Skills pathways, multi-family data isolation. Each evidence record runs through a three-call Claude pipeline. Before anything writes to coverage, a guardian approves, adjusts or rejects what the pipeline suggests. Built solo, end to end.
A procurement training program rebuilt as a knowledge system, not a course. Phase 1, already shipped, moved the platform from WordPress to a custom Next.js, Supabase, Stripe and Vercel stack. An AI-assisted production pipeline cut triage-to-first-draft time by about 55 per cent. Phase 2, now in design, treats the procurement corpus as a knowledge layer with a contract. Instead of 18 hours of material to wade through, the learner's real tender becomes the query. The system scaffolds novices and runs formative assessment over their actual work.
If you're working on AI-assisted training, outcome mapping or content operations at scale, I'd be happy to compare notes. I'm open to permanent AI–L&D roles in Australia and to consulting work through Duvanced.