Policy-gated AI
Approved templates can move automatically; risky or low-confidence work stays queued for review.
Selected project
OpsDesk AI
A policy-gated service desk demo that turns customer enquiries into reviewed operational work.
The project shows a practical AI workflow for local businesses: capture the enquiry, extract intent and risk, route approved templates, and keep complex cases in human hands.
Operational record
The desk keeps raw enquiry text, AI summaries, selected templates, confidence, tokens, and cost visible.
Workflow demo
A full path from enquiry capture to triage, template routing, watchtower events, webhooks, and architecture notes.
Developer tooling
Ralph Loop
A small TypeScript CLI for draining a folder of reviewed Markdown implementation plans through Codex, one focused run at a time.
$ ralph --repo ~/work/app --plans ~/plans/queue
- 01 Pick next runnable *.plan.md
- 02 Start Codex with one focused plan
- 03 Run simplify/review after completion
- 04 Archive completed plans into done/
reviewed plans -> fresh Codex context -> verified change ->
next plan
What it demonstrates
AI product thinking with production-minded workflow design.
- Service-desk workflow modeled around human approval, not blind automation.
- Product UI designed for operators: dense, inspectable, and status-oriented.
- AI tooling designed around plain files, fresh context, and auditable runs.