/ Insights
Written up as architecture case studies: the context, the decisions I owned, the trade-offs behind them and the evidence that came out. Shorter engineering notes cover narrower pieces of the same work.
/ Case studies
Internal platform · Paved path
I built the Dev Harness: an internal paved path for enterprise AI delivery. It packages reusable application templates, deployment automation, MCP and tool-integration patterns, security guardrails, evaluation workflows, observability and CI/CD so an AI application starts from a supportable baseline instead of an empty repository.
Retrieval architecture · ReAct agents
Architected and delivered a scalable enterprise RAG and agent platform: a 500GB+ embedding and ingestion pipeline feeding hybrid search with reranking and vector retrieval, behind a ReAct agent architecture. Retrieval quality was treated as the product — measured chatbot accuracy improved to 95% and response latency was reduced.
ML platform · 100+ AWS accounts
Designed and implemented a secure ML and model-serving platform spanning 100+ AWS accounts. Open-source HuggingFace models are served through Amazon SageMaker with GPU-backed inference, automated model packaging pipelines and centralised governance controls — so teams keep their own accounts while security and lifecycle standards stay central.
Grounded assistant · Procurement
Designed and productionised a grounded enterprise Procurement assistant on Azure AI Foundry, using LangGraph for orchestration, ChatKit for the interface, PGVector for retrieval and integrations with MCP servers. Answers stay grounded in trusted enterprise documents, and the architecture was standardised into a reusable reference implementation.
Operating model · Platform standards
Designed the operating model that lets generative AI scale beyond pilots: identity lifecycle and SCIM provisioning, RBAC, a ServiceNow approval workflow, usage controls, licensing and cost governance, plus the ADRs and platform standards that guide technology selection across engineering teams.
/ Engineering notes
Engineering note · Synthetic load test
An engineering note on a controlled synthetic load test against an AI Gateway. More than 69,000 requests were generated by the test harness — synthetic traffic, not production traffic — to exercise the gateway path under sustained load. Framework and platform evaluation was run separately, as a qualitative architecture comparison.
/ Next step
Open to full-time Senior–Staff AI / agent platform roles (remote-friendly), as well as contract and consulting engagements. Response within two business days.