Revenue Integrity & Coding Platform
Software that checks each procedure a medical practice performed against what insurance actually paid, flags what was missed, and codes the practice’s hospital work with AI that a human coder approves.
Today, without it:
- visits in one system
- payments in another
- paperwork read by hand
- unpaid claims unnoticed
With the platform:
- Clinic visits & payments (Exported each month)
- Hospital paperwork (Scanned documents)
- Codes suggested (by AI; Read, sorted, and coded)
- Coder approves (Every suggestion reviewed)
- Each procedure matched (To what insurance paid)
- Revenue, claim by claim (By procedure and insurer)
- If not paid: Unpaid or $0 claims (Each in its own queue)
- Auditor resolves (Notes kept on every re-upload)
The hard part: AI suggests the codes; a human coder decides.
The problem
A specialty practice’s money moves through systems that do not talk to each other: its records system logs the visits, a billing company reports the payments, and hospital work arrives as scanned documents. A procedure that was never paid can go unnoticed, and coding hospital work means reading every document by hand.
What it does
- Each procedure checked against what insurance actually paid
- Unpaid and $0 claims put in front of an auditor
- Hospital paperwork coded with AI, then approved by a human coder
- A complete billing packet filed and emailed in one click
- Revenue by procedure and insurer, down to each claim
Matching that never guesses
Each procedure gets the same deterministic identity on both sides — patient, date of service, and procedure code — so reconciling the practice’s records with its payments is an exact lookup, not fuzzy matching. Data is partitioned by month, re-uploading an export never duplicates rows or overwrites an auditor’s notes, and a $0 payment is kept distinct from a missing one, so each lands in its own audit queue.
From scanned PDF to coded billing packet
Google Document AI reads each hospital document, and a fine-tuned Llama 3 70B, run at temperature 0, classifies it, extracts the encounter details, and suggests ICD-10 or CPT codes. A coder reviews every suggestion, with ICD-10 search built in from the U.S. National Library of Medicine. One click then assembles the facesheet, coding sheet, and notes into a single packet, files it in Google Drive, and emails it to the billing company. Each coded procedure becomes a claim row, so hospital work is audited exactly like clinic visits.
Every number traces back to its claims
- Revenue trends with a three-month forecast whose confidence is explained, not hidden
- Procedure-code profitability and insurance payer rankings
- Accounts-receivable ageing, computed once when payments are uploaded
- Drill-down from nearly every chart to the claims behind it, with CSV export
Architecture
- A React application on Firebase Hosting
- 35 Cloud Functions doing all parsing, matching, analytics, and AI work
- Four roles and sixteen permissions, checked on the server for every privileged action
- Google Workspace sign-in, with the Google Drive and Gmail APIs
The diagram on this page is an illustrative system flow, not a screenshot of the product.