Drop a start-of-care referral packet in. Six minutes later your coder has the primary and other ICD-10-CM diagnoses, the OASIS answers the packet supports, and a quote from the page for every code. Review, sign, bill.
A start-of-care packet is 10 to 30 faxed pages from three different EHRs. The diagnoses that set the agency's Medicare payment are in there, next to patient questionnaires, resolved conditions, outdated codes from the referring office, and contradictions nobody flagged. A certified coder spends most of the next half hour looking, not deciding.
Cover, order, discharge summary, H&P, med rec, labs, questionnaire, PT eval. Three of the pages are upside down.
The questionnaire says “Yes” to COPD. The H&P says never smoker. The order says N18.3, which stopped being a valid code in 2020.
Hypertension plus heart failure plus CKD is a combination code. Miss it and the claim is wrong.
At 30 minutes a chart, one coder clears 16 a day. At $25 to $55 a chart, agencies pay for every one of those minutes.
Everything that can be checked without a model is checked without one. That's why the output doesn't need re-verifying.
Faxes, rotated scans, handwriting, three EHR layouts in one packet. Each finding is recorded with its page, a verbatim quote, who documented it (physician, patient form, lab) and whether it's active, resolved, suspected or denied.
The full OASIS-E2 guidance is in context on every case, not a summary. The coder's own rules: primary is the chief reason for care, only provider-documented diagnoses count, resolved and rule-out conditions stay out.
Every code is looked up in the official ICD-10-CM table: does it exist, is it reportable, is it allowed as a primary under PDGM, is a symptom-control rating permitted. Anything that fails goes back for correction before you see it.
A worksheet with the codes, the evidence, the conflicts, the provider queries and the OASIS answers the packet supports. Accept, edit, sign. Every edit is logged, and every edit makes the next packet better.
Move the sliders to your own numbers. Oasai is priced at $6 per start-of-care chart; the review time below is what pilot coders averaged with a finished worksheet in front of them.
Assumes Oasai at $6 per chart and no change in headcount. Freed hours become capacity for more clients, OASIS review, or ADR work. Your numbers will differ; the pilot measures them on your own packets.
This is a real worksheet from a pilot packet: a 79-year-old discharged after a heart failure exacerbation, with diabetes, kidney disease and atrial fibrillation. Names are synthetic. Everything else is what the coder saw.
Hypertension, heart failure and CKD together require the combination code I13.0 first. The referring office had written I50.9 and I10.
E11.9 became E11.22 and the invalid N18.3 became N18.32 because the chart documented diabetic CKD at stage 3b.
“Rheumatoid arthritis: Y” on a patient checklist with negative RF and CCP: not coded, with the manual page cited. Resolved AKI: excluded. “Possible OSA”: excluded.
Inpatient stay, active diagnoses, height and weight, dyspnea, continence, wounds, high-risk drugs, oxygen: 10 of 10 matched the answer key.
| What you get | Human-only coding | Generic AI suggestions | Oasai |
|---|---|---|---|
| Time per chart | 20–40 min | 15–30 min (re-verifying) | 5–10 min |
| Evidence per code | Coder's memory | None, or a confidence score | Page and verbatim quote |
| Codes checked against the official table | Manually | No | Every code, every time |
| Patient checklists vs provider notes | Coder judges | Often confused | Separated at read time |
| OASIS cross-check | Separate review | Not covered | Included |
| Provider queries | Written by hand | Not covered | Drafted, ready to send |
| Cost per chart | $25–55 outsourced; $10–20 in-house labor | Subscription | $6 |
“I stopped trusting AI coding tools two vendors ago. This one shows me the page. If it's wrong, I can see why in ten seconds, and it's wrong a lot less than I expected.”
“Our average went from 31 minutes a chart to 9 in the first month. Same five coders, two new agency contracts, no overtime. The savings calculator on this page was, if anything, conservative for us.”
“The provider queries alone were worth it. It catches the obesity-versus-BMI problem and the stale problem-list entries that used to come back to us in ADRs.”
Runs in AWS under a Business Associate Agreement, with model inference through Amazon Bedrock under BAA. Encrypted at rest and in transit, access logged, retention set per customer.
Page, quote, the manual guidance it rests on, who reviewed it, what they changed and when. That's what an ADR asks for.
Oasai hands your coder a worksheet, not a verdict. Accept, edit or reject each code; the decision and the signature stay with your certified staff.
Every code is checked against the NLM ICD-10-CM table and the PDGM primary-diagnosis rules before you see it. No invented codes, no retired codes.
Published only as measured against certified coders on held-out real cases, with the method stated. No marketing percentages.
Packets deleted 30 days after a case closes by default. Corrections improve the shared model only with your written agreement, after de-identification.
Oasai is built by NextVibe Innovations, a Barcelona startup focused on AI-powered automation for US home healthcare. Before Oasai, the team built the document pipelines that read referral packets for hospital systems and revenue-cycle companies: the same faxes, the same EHR exports, the same contradictions. Home health coding was the obvious next job for them.
Built and ran document-intelligence products for regulated industries before founding NextVibe. Owns the customer relationships and the accuracy program.
Senior data engineer behind healthcare referral extraction pipelines on AWS and Bedrock; designed Oasai's read-code-check architecture and the evidence model.
Working home health coders who built the regression set and decide what “correct” means. Nothing ships that they would not sign.
No. It replaces the 20 minutes of reading before the decision. Your certified coder reviews, edits and signs every case; Oasai only prepares the worksheet. Teams use the freed hours for more clients, OASIS review and ADR work.
On pilot packets with independent answer keys, Oasai selected the correct primary and matched 10 of 11 secondary codes, with every exclusion correct and 10 of 10 OASIS cross-checks matched. We publish accuracy only as measured against certified coders on held-out real cases, with the method stated, and the free demo lets you measure it on your own packets before you pay anything.
The packet, as a PDF or TIFF, the way it arrives by fax. No EMR integration is required to start; drop files into the console or send them through the API. Results export in the formats your coders already paste into.
Oasai runs on AWS under a Business Associate Agreement with model inference through Amazon Bedrock under BAA. We sign a BAA with each customer. Packets are deleted 30 days after a case closes by default.
$6 per start-of-care chart and $4 per recertification or resumption, with volume discounts from 5,000 charts a month. Firms that run their coders through the console pay a $1,500 monthly platform fee for seats, edit logging, the accuracy dashboard and the API. The first 10 charts are free in the demo.
Home health start of care, recertification and resumption today. Hospice coding is in pilot with two firms; ask if you want to be the third.
Request a key, open the demo, drop in 10 de-identified packets your coders have already coded. Each worksheet comes back in minutes, straight from the system, with no one in between. Compare it against your own answers; if it doesn't save your coders time, you've lost an afternoon.
Fill in the form. The key arrives by email within one business day.
Several at once, in parallel. A 15-page packet takes 3 to 6 minutes.
Codes, evidence, conflicts and OASIS answers next to what your coders decided. Then you decide.
Your demo key arrives by email within one business day, good for 10 charts. Paste it at oasai.truecircle.com/demo and drop your packets in. If you'd rather talk first, reply to that email and we'll set up 20 minutes.