Production-grade SaaS, AI agent infrastructure, edge ML on mobile, and two systems in draft. Read for outcomes — what shipped, where it lives, what it does — or read for architecture — what the topology looks like and how to walk it.
Production FinOps SaaS — cross-account AWS scanning, CUR 2.0 attribution, Bedrock-streamed remediation.
Live SaaS · cloudwastehunter.io
Cross-account AWS scanning over STS AssumeRole. Cost attribution via Glue + Athena over CUR 2.0. Findings enriched with Claude Haiku on Bedrock — the model explains the risk and emits the exact Terraform fix.
Live SaaS · cloudwastehunter.io
Decoupled Next.js + AWS Lambda monorepo. Anthropic Claude API behind a strict master system prompt that enforces structural completeness. The brief is copy-pasteable into Claude Code, Cursor, or Windsurf.
Live · vibingcoderai.com
Google ML Kit tracks 33 body landmarks at 30 fps; joint-angle math evaluates rep quality and triggers corrective audio cues. Supabase + RevenueCat behind it. The best cloud architecture is sometimes knowing when not to use the cloud.
30 fps · on-device pose detection
Multimodal model reads photos + symptom descriptions against a structured triage framework. Output is a prioritised recommendation grounded in observable signs — "monitor at home" vs "go to the clinic now".
Development started · walkthrough drafting
Medication adherence, appointment coordination, insurance document management, family role-based access. The daily briefing surfaces issues before they become crises — "Mum's BP was high yesterday; her cardiology appointment is in 3 days".
Concept phase · architecture drafting
How I work
Production-first systems, infrastructure as code, cost-aware architecture, AI as leverage — never as substitute. The five projects above are what those principles look like in shipped form.
Full principles