15 years across embedded Linux edge systems and AWS cloud infrastructure. End-to-end platform ownership spanning C/C++ devices deployed in 120+ countries to a multi-account AWS environment serving millions of requests per day.
A real-time prompting system for executives and speakers — transcribes the conversation live, retrieves the exact fact you need from your own knowledge base, and surfaces it on your screen within half a second so you respond calmly instead of freezing.
Proof of concept built in 2 days. Zero-latency teleprompter on a single Mac. Apple Silicon NPU transcription, vector-semantic RAG (pure numpy, no external DB), and a multi-role web UI where staff inject tactical cues into the speaker's display in under 0.5s.
Enterprise rewrite with a cross-platform C++ inference engine (whisper.cpp, gRPC), Go BFF gateway, and React frontend — dual-mode architecture that runs identically on local LAN or cloud EKS. 21 Architecture Decision Records document every trade-off.
Multi-account AWS landing zone built from zero — Organizations with SCPs, Identity Center, GitHub OIDC federation, Terraform with native state locking, EKS with ArgoCD GitOps, and Prometheus/Grafana observability. 17 Architecture Decision Records. Fork-and-deploy from a single YAML config.