Field Compliance Desk

Wrenfield Home Care

Independent prototype of Ultimate Care's caregiver compliance audit, by The AI Pipe
Invented data, live agent

A renewal reminder looks forward. The days already behind it are the ones an audit reads.

Fourteen personnel files at an invented New York agency, eight required items on each, and a visit ledger. The agent reads every file item by item and says what each document shows. The code does the rest, and it writes every word on every card below. Independent prototype by The AI Pipe on invented data, not affiliated with or commissioned by the agency named above.

The case this desk exists for

Computed, no model call

Scenario date 2026-09-16. Read off a run recorded on this page, which took a moment to read all fourteen files, produce the state of every item, the documentary results, the four deadlines and the cross check against the visit ledger. It did not write the monthly report or the escalation list, which are the fourth button. Every question below shows its own time under its own answer.

Fictional desk. Do not paste real personnel, employment or patient records. What you type goes to the model to answer the question. It is not kept in our statistics and nothing on this page is stored against you.

Enter to send, shift and enter for a new line. Live tool use over fourteen files, two procedures and six public rules.

Take this with you

A selection of the personnel file audit checks a New York licensed home care services agency works under, every line carrying the passage it comes from and the URL that passage was read from, and saying whether this desk tests it, names it, or leaves it outside. Built from the public text quoted on this page. Not the list of everything a file has to carry. Not legal advice.

See the lines

Files on the desk

Procedures and public rules

What this shows

The problem: A personnel file is a stack of documents that arrive at different times and fall due on different clocks. Somebody has to read each one the same way every time, know which figure applies to which title, notice the file where a document fell due before a visit that has already gone out on an invoice, and get the question to the right person before a surveyor asks it first.
The result: The run at the top of this page is real and is recorded: fourteen files read item by item, every state and documentary result computed from the dates on the records, the four deadlines counted in business days, and every line carrying the passage it came from.
Where the words come from: Every word on a desk card is written by the code from the record. The model reads each item off its own section, chooses which open point leads a row and which rule to quote, and there is no field on any card for a sentence of its own: the headline, the summary, the next step, the questions and every count are composed here. A quotation is checked word for word against the quotation field of the passage, never against the commentary printed around it, and scripts/verify_quotes.mjs rechecks all forty three of them against the copy of the official page they were read from. What the model writes in its own voice sits above the cards, labelled as its commentary, and the desk runs its word checks over it and shows what they find rather than claiming the sentence was impossible to write.
What it does of what the postings describe, and what it does not: The postings ask for compliance audits of caregiver records, tracking of certifications, CHRC, HHA documents and medicals, confirming caregivers before an operational handoff, and escalation to leadership. This prototype audits documents already indexed on a file, tracks the intervals and the deadlines, compares the windows a file covers with the visits already billed, and produces the escalation list with the person each line goes to. It opens no scanned document and extracts nothing from an image, which is the scanning and indexing the first posting puts at 30 per cent of the role. It touches HHA Exchange, E-Verify, PaperFlow, PaperVision, SharePoint and the Health Commerce System not at all. And it produces no authorisation to begin an assignment: it produces a documentary result over five items, and FCP-110:1.5 leaves the decision to a named person.
What is simulated, and what is not covered: The agency, its three offices, the fourteen aides, their clients, every document date and the two internal procedures are invented. The six public rules are not: 10 NYCRR 766.11, Part 402, Part 403, the state guidance letter DHCBS 22-07, the USCIS handbook M-274 and the E-Verify process are quoted verbatim, each passage in its own field, printed with the URL it was read from and the date it was read, 2026-09-17. There is no connection to any system, no patient information of any kind, and no scanned document. This is not a hiring tool and it makes no hiring decision. It never states that anybody is or is not authorised to work, never proposes any action about an employee, never reads what a criminal history determination contains, and never says that a service was improper, that a claim is or is not payable, or what any survey or accreditation review would find. Accreditation standards are not public text, so no wording of any of them is on this page. Nothing here is legal advice.
What a pilot would need from your side: Your blank personnel file checklist, your own policy on renewal intervals and warning windows in whatever form it exists today, the column headings your visit records already produce rather than the records themselves, a short conversation with whoever runs compliance and one person who can say whether a reading is right, and counts rather than files: how many aides fall under each document type, how many renewals come due in a month. Aides, their documents and their clients stay invented until your legal and security leads have agreed, in writing, which fields a pilot may see, who holds them, where they sit, how long they are kept and how they are deleted. No personnel file is needed to start, and removing names is not anonymisation.
Who built it: Frederic de Lavenne de Choulot, The AI Pipe. Built as a demonstration on invented data, not sold as a compliance, legal or employment product.