Your seven AI ideas, built into working agents — not a slide deck.
You sent PRR seven productivity ideas. Rather than describe them, we stood each one up as a clickable demo on the same production stack we run for HSI, HHS, USF, and Marsh McLennan. Every screen shows the human-in-the-loop pattern we'd deploy at Carnegie.
Each idea, mapped to a working screen
Straight from your email to us. Click any card to open its live demo.
Receipt & expense automation
An MCP server that reads program directors' expense spreadsheets, extracts every line, maps GL codes, and stages Concur-ready entries for a human to approve — instead of coordinators re-typing.
Open demoLibrary research across journals
Agents that use the library's shared journal credentials to search, pull, and summarize articles across many third-party databases at once — so scholars stop logging into each site separately.
Open demoContact enrichment & dedup
Agents that continuously maintain the contact database — detecting duplicates, merging records, and enriching titles, orgs, and topics — through a Salesforce connector.
Open demoChina AI Initiative tracker
A knowledge-management system where background agents monitor sources, extract China AI initiatives, place them on a map with drill-through data, and keep it current. Same pattern for the Climate team's DDD project.
Open demoBrand & content QC monitor
An agent that periodically scans the website, publications, and social posts for broken links, off-brand assets, and factual inconsistencies — and files fixes.
Open demoEvent logistics copilot
Drafts invitee communications, tracks RSVPs, chases non-responders, and keeps every deadline moving — bridging Marketo today and whatever tool you choose next.
Open demoAsk-Andy intranet agent
Staff ask Claude and get reliable answers grounded in the Comms and HR documents on Andy — branding, onboarding, policy — and Claude returns the actual source document.
Open demoOur recommendation: start with one high-leverage pilot, architected for all seven.
Several of these share the same backbone — an MCP + agent layer, secure connectors, and a human review step. We'd land one pilot fast (the Concur receipt agent and the China AI KMS are strong candidates), prove the value, then reuse the platform for the rest.
See the engagement plan