Clockwork
An agentic assistant for residential real-estate teams. Every morning it reads the book of business and drafts the day — follow-ups, listing updates, market notes — as a queue a person reviews. The agent proposes, a human approves, the agent sends.
One demo run composes about 150 actions.
VitalPoint
Scores every account in the book on usage, support load, and customer sentiment. The Renewal Desk ranks the quarter’s conversations, and each at-risk account gets an evidence-backed brief — tickets, sentiment, usage — that a rep can act on the same day.
ZKicks
An agentic chatbot over five years of QuickBooks-style accounting data for a fictional sneaker reseller. Ask it revenue by brand and it writes its own SQL against a real double-entry ledger, renders the table and the chart, and searches the company’s emails and memos with citations. Ask what the data can’t answer, and it abstains — and says why.
An eval suite runs in CI, including an abstention eval most retrieval demos never test. Now being productized as a multi-tenant, white-label financial chatbot with row-level-security tenant isolation.
TrendHUD
Longitudinal lab visualization for physicians — every coded lab from an EHR export on one canvas, trended over years instead of one visit at a time. FHIR-native, LOINC-coded, and live at trendhud.app.
TrendME
The patient-side sibling of TrendHUD — your own labs, your own trends, in a form you can actually read. No screenshots yet; the wordmark arrived first.
FitChecks
An AI stylist. Photograph the outfit and FitChecks scores it — trend, color, fit, style, each on its own — writes a real style assessment, and hands you a card worth sharing.
