A handful of public repos — the showable tip of the iceberg. Small sample, but they run, and they're the kind of thing I build everywhere. Full profile on GitHub.
Sharing the thing you just made — a report, a dashboard, a one-off analysis — usually means mangling it into an attachment nobody can open properly. PageVault publishes it to a real URL instead, straight from Claude, ChatGPT, or any MCP-capable tool, and lets you decide who can open it. Runs on your own Cloudflare account: unguessable public links, or email-gated documents and per-client portals behind Zero Trust.
Serverless AWS cost monitoring with AI-driven anomaly detection and analysis of what's actually moving your bill — pushed to Slack, with a bot you can @mention for ad-hoc questions. CDK-deployed, about $5 a month to run. The cost discipline I want on every project, automated.
The plumbing behind good RAG: a Python framework for populating vector databases from messy real-world sources — websites, PDFs, docs, cloud storage — with the chunking and filtering that decide whether retrieval actually works. Pinecone or Milvus.
A focused tool for turning raw source artifacts into clean, structured context an AI can actually use — the front of the vector-search and RAG pipeline, where retrieval quality is usually won or lost. Pairs with the loader above.
Self-hosted Matomo analytics on AWS — own your data and cut the marketing-tech bill, wired into Shopify and a wider data warehouse. Deployment scripts included.
A DevContainer template that sandboxes Claude Code (or any AI assistant) away from your host — so you can give an agent real access without handing it the keys to everything. The security layer I run my own vibe-coding through.