Self-built demonstration projects using fictional companies and synthetic data. No client data is shown.
Classifies every customer email, drafts replies from a policy sheet, quarantines phishing and prompt-injection attempts, escalates legal threats, and sends nothing without human approval.
Result in testing: a fake $4,800 refund email with hidden instructions to the AI was quarantined with no draft.
Three pages, a star-schema model, 20+ DAX measures and row-level security by region, for a fictional outdoor-gear company (about $22M in sales, 36k order lines). It surfaces the margin erosion caused by discounting in one region.
Reads job-alert emails, scores each job 0–10 with reasons and red flags, drafts a proposal only for strong fits, and sends a ranked daily digest. It runs on a private server with zero inbound ports and phone alerts on failure.