OCR + AI extraction with document classification, ICR learning, KFI review, and full data warehousing. Processes checks, EOBs, remittance advice, and claim letters.
Checks · EOBs · Remittance · Claim Letters
Side-by-side PDF and form for QA corrections. Low-confidence fields are flagged automatically.
Persistent record store across all sessions. Search, filter, sort, and export your full history.
Insights across all extracted and warehoused records.
Intelligent Character Recognition — the system learns from your corrections to improve future extractions.
When you draw a region on the PDF and label it, the system remembers where each field is located for that insurer's documents. Future extractions get these location hints.
Everything MedCheck Pro does, in plain English.
Along the way, 🧠 ICR Learning watches your corrections and remembers them, so the same mistake doesn't need fixing twice.
Extraction runs in phases: first every page is OCR-scanned and classified (is it a check? an EOB? a letter?), then pages are grouped into per-patient records, and finally the AI pulls your selected fields from each group.
The results table shows one row per payment record. Any cell can be edited inline — just click it. The Pause button halts between pages if you need to step away mid-batch.
When a check combines multiple payments or includes interest (e.g. principal + INT $15.12), the breakdown is captured in the Payment Breakdown column so the total on the check is always explainable.
Certain — read directly from machine text or MICR line. High — clearly printed, high OCR quality. Medium — legible but worth a glance. Low — smudged, handwritten, or ambiguous. Always review these.
Records with any low-confidence field are automatically added to the KFI Review queue — the count appears as a badge on the KFI Review tab.
Open any record to see the original PDF page beside the extracted fields. Fix any value and hit ✓ Save Corrections — or ✓ Approve as-is if it all checks out. Arrow keys flip between pages; Esc closes.
Teach it layouts: use the drawing tools (▪ Rectangle / ✏ Freehand) to circle where a field lives on that insurer's documents, then label it. Future extractions from the same insurer get those location hints — this is what the "Check Layout Templates" memory on the ICR tab stores.
Every correction you make in KFI Review is remembered as a rule: "when the AI reads X for this field at this insurer, it actually means Y." The next extraction applies your rules automatically. The ICR tab lists every learned rule, how often it has been applied, and lets you delete ones you no longer want.
Hitting Save Session after an extraction stores the batch here, grouped by save date. Search by name, insurer, claim or check number; filter by date, insurer, doc type, or status; and export any slice to CSV.
Re-saving is safe. If you correct a record in KFI Review and save again, the warehouse copy is updated — not duplicated. Each record carries a permanent internal ID for exactly this reason.
Two records are treated as the same payment when all three of these match: the check number (ignoring formatting — "N 255397352" and "255397352" are the same check), the patient name, and the amount. If a record has no check number, the check date is used instead, so unrelated payments are never merged.
Re-running the same file never creates duplicates. If you extract the same PDF again, matching records simply update the existing warehouse rows. A duplicate flag only appears when the same payment arrives from a different file — e.g. a check that was scanned into two separate batches.
Flagged a pile of them from earlier testing? 🧹 Remove Duplicates in the warehouse toolbar clears every flagged copy at once — the original of each payment always stays.
When a duplicate is detected, it stays visible in the warehouse with an orange ⚠ Duplicate badge and is excluded from every total — the stats tell you exactly how many were excluded, so the numbers always add up.
Analytics combines your warehouse with the current unsaved session, counting each unique payment exactly once. Flagged duplicates are excluded (until you Keep them), so the Total Records and Total Amount here always agree with the Data Warehouse stats.
You get amounts by insurer, document type counts, a recent-activity feed, and your ICR correction rate at a glance.
MedCheck works fully without an account — your warehouse and learned rules live in this browser. Signing in (when available via the Sign In button in the header) syncs everything to the cloud so it follows you across devices. Anything you did as a guest is merged in on first sign-in, never lost. The header indicator shows Local only, Syncing…, or Synced.
Why don't my warehouse totals match my last extraction?
The warehouse holds all sessions, not just the last one — and excludes flagged duplicates. Check the orange duplicates card in the warehouse stats: it states exactly how many records aren't counted and each one can be reviewed with Compare.
Can I close the tab during extraction?
Minimizing or switching tabs is fine — processing continues. Fully closing the tab stops the run; use Pause first if you need to.
Where is my data stored?
In your browser (localStorage) by default; in your private cloud space as well if you sign in. Export CSV any time for a portable copy.
A field keeps extracting wrong for one insurer.
Fix it once in KFI Review — ICR learns the correction. For stubborn layouts, draw a region template on the PDF so future extractions know exactly where to look.
Pay once per batch. No subscriptions, no surprises.
Payments processed securely by Stripe. Credits never expire. Questions? support@medcheckpro.com