Healthcare receivables agency
≈ $10M revenue (approx.) · ~60 employees · regulated healthcare receivables
Protected-health-information send gate
Every outbound file that carries protected health information is checked against the placement it belongs to and held for human approval, with an audit trail of who released what, and when.
- Organization
- Healthcare receivables agency, on-premise, regulated
- Risk
- Outbound files carrying protected health information
- Runs on
- The agency's own hardware
- Scope
- Send verification with human approval and audit
Where they started.
A file sent to the wrong recipient with protected health information inside is the agency's worst day. The manual control was attention: a person checking a highlighted list before hitting send. It worked until the day it did not, and the compliance team that followed needed a control that could not be skipped.
What was built.
- Built on FlatClaw's approval engine: the agent composes every outbound send that carries protected information, but never executes it.
- Before the file reaches the queue, the agent checks it against the placement it belongs to, the recipient, and the expected shape of the data, and attaches the evidence and any anomalies.
- A compliance reviewer approves or denies from the queue; on approve, the send replays with the reviewer's own credentials; on deny, nothing happens.
- Both outcomes, with the approver's identity, land in the audit log.
What changed.
- The control is structural: a send cannot happen without a recorded human decision.
- Reviewers see evidence, not a highlighted spreadsheet.
- Audit questions are answered from the log, not from memory.
- The same engine gates every other consequential action the agent can take.
The files being checked are the sensitive data itself. Checking them with a hosted model would create the very exposure the control exists to prevent.
On-premise hardware
Components involved.
More like this.
Healthcare receivables agency
≈ $10M revenue (approx.) · ~60 employees · national client base
- Use Case:
- Placement-file intake automation
- Results:
- Placement files arriving in any layout are normalized into the agency's existing upload format, with a human approval queue before anything reaches the system of record. A one-and-a-half-person manual job becomes a review step.
Sleep-therapy provider
≈ $60M revenue (approx.) · 40 clinics · one clinical reporting app
- Use Case:
- Natural-language clinical reports
- Results:
- Clinicians ask for patient cohorts in plain English and a constrained pipeline turns the question into validated filters, never free-form database queries, on a private GPU that costs a fraction of the hosted model it replaced.
Global law firm
≈ $3B+ revenue · Thousands of lawyers · dozens of offices
- Use Case:
- Walled-matter research assistant
- Results:
- Document search and drafting inside the firm's own tenancy: every answer cited to its source, every matter isolated by role, and nothing ever sent to a third-party model.
Have one like it?
Every spotlight started as a conversation about a process nobody liked doing, under a data-locality constraint.