AI Back Office Automation for Home Health Provider
A home health provider was processing physician orders and CMS forms by hand. The agentic system we built cut order processing from 5 minutes to 30 seconds at 90%+ OCR accuracy across 20+ document types. Their own words: ‘We’re looking at $250,000 in savings per year, and that’s just one of the projects.
What this engagement became
Back Office Bottlenecks
Before the System Existed.
Problem 1
Manual fax processing
Every incoming physician order — CMS 485 forms, recertification orders, therapy evaluations, start-of-care documents — had to be printed, filed by hand, and entered into the EMR manually. Each order took up to 20 minutes and left room for errors at every step.
problem 2
Delayed patient care
Orders sat unprocessed for hours, sometimes days. That lag pushed back patient start-of-care dates, hurt quality scores, and stalled episode of care progression across the board.
Problem 3
Compliance and audit gaps
There was no digital trail for incoming orders. When physician signatures were missing, the only option was to resend and wait — adding days to an already slow process and putting RAP payments at risk.
Problem 4
15+ hours lost every week
Staff were spending the bulk of their time on fax handling and data entry instead of clinical work. There was no capacity to scale without adding headcount.
Problem 5
No single view of order status
Orders lived across physical files, email inboxes, and fax systems with no unified view. Tracking episode status, spotting missing OASIS signatures, or pulling a compliance report required manual digging every time.
Solution Planning
Agentic System Architecture
Webhook-triggered intake
Every fax enters the system the moment it arrives via RingCentral webhooks, queued for concurrent processing without manual handoff.
AI-powered classification and extraction
AWS Textract classifies each document across 20+ types and extracts structured patient data — MRN, DOB, insurance, start-of-care dates — with confidence scoring and fallback logic built in.
Intelligent signature validation
Computer vision detects and validates both ink and digital physician signatures before any document moves forward. Unsigned orders are flagged and re-routed automatically.
Automated EMR upload
Playwright navigates the healthcare platform, finds the right patient record, and uploads to the correct task type — no API access needed, no manual steps involved.
Full audit trail
Every action across the pipeline is logged in PostgreSQL with timestamp, actor, and document reference. Slack alerts fire on any failure. Nothing moves silently.
Centralized monitoring dashboard
All automation tasks across faxing, OCR, uploads, and sync are tracked in real time with success metrics, error categorization, and performance analytics.
Solutions We Provided
An Agentic System Built for
End-to-End Order Automation
Automatically Receiving and Processing Faxes
RingCentral webhooks instantly receive and queue incoming faxes for OCR extraction—eliminating manual downloads. The system processes multiple faxes concurrently (including CMS 485, CMS 495, physician orders, PT/OT/ST evaluations, and recertification orders), maintains a complete audit trail, and updates status in real time.
AI-Powered OCR Document Extraction
Using AWS Textract, the system automatically extracts key patient and order details (MRN, DOB, start-of-care date, insurance information, physician details, order numbers) from 20+ document types with 90%+ accuracy—removing manual data entry and reducing errors by 95%. The OCR pipeline classifies document types including CMS 485 Plan of Care forms, therapy evaluations (PT, OT, ST), recertification orders, and frequency discontinue orders.
Fax Receipt and Queue Management
RingCentral webhooks trigger processing the moment a fax arrives. An asyncio-based queue handles concurrent faxes without delays, bottlenecks, or manual handoff at any point.
AI-Powered OCR and Data Extraction
AWS Textract classifies each document across 20+ types and pulls structured patient data — MRN, DOB, insurance, start-of-care dates — with confidence scoring and automatic retries on failed jobs.
Physician Signature Detection
BAA on every engagement. US-region AWS deployment. No cross-border PHI movement. NDA standard, not optional.
EMR Document Upload
Playwright logs into the healthcare platform, finds the patient record by MRN or name, and uploads to the correct task type — EOE Ready, OASIS, or PT Fax — with no API access or manual steps required.
Outbound Fax to Physicians
Pending orders are retrieved, sent to physicians via RingCentral or HumbleFax APIs, and tracked through to delivery confirmation. Order turnaround cut from days to hours.
Centralized Logging and Alerts
Every task across faxing, OCR, uploads, and sync is logged with success metrics and error tracking. Slack alerts fire on failures. Nothing moves silently through the system.
What the Client Said · Clutch
I used to have a five-person team whose job was to read and send out faxes. Now I have half a person assigned to that. We’re looking at $250,000 in savings per year — and that’s just one of the projects. The good part is I don’t have to scale up or down my team because the automation is running so efficiently.
Client
End-to-end workflow
Where the Technical Challenges Were
and How We Solved Them
From a sentiment-driven Designer’s Choice browse to a recipient scanning a QR card and
watching a personal video — one ordering flow, six automated steps, AI-assisted at every relevant beat.
01
Converting Fax Images to Structured Data
Physician orders faxes varied in layout, quality, and content, making data extraction difficult. We integrated AWS Textract with custom pattern matching for 20+ document types and handled async jobs via SNS/SQS—achieving over 90% accurate structured data extraction.
02
Handling Concurrent Incoming Faxes
Multiple faxes arriving together caused processing delays and data conflicts. We implemented an asyncio-based queue with background workers, optimistic locking, and retry mechanisms—enabling smooth, parallel OCR processing of three or more faxes simultaneously without blocking or performance loss.
03
Detecting Physician Signatures in Noisy Faxes
Low-quality faxes with manual and digital signatures complicated verification. Using AWS Textract, OpenCV, and PyPDF2, we identified “Physician Signature” regions, validated authenticity, and cropped signature areas—achieving more than 95% accuracy across diverse fax formats and layouts.
04
Ensuring Data Integrity and Audit Compliance
Healthcare regulations required full traceability of every faxed order. We created a detailed database schema with OCR logs, automation records, and JSON audit trails—ensuring complete HIPAA-compliant visibility from fax receipt to EMR upload with guaranteed data integrity.
05
Automating Fax Sending with Dynamic Order Retrieval
Manual fax sending from multiple EMR tabs (PT,OT, ST, MSW, OASIS, )was time-consuming and error-prone. Using Playwright, we automated navigation, pagination, and fax dispatch via APIs—achieving 95% automation success and reducing order processing time from five minutes to just thirty seconds.
How We Built It
Our Full-Cycle Automation
Development Approach
Every phase connects directly into the next. Webhook intake feeds the OCR pipeline,
extraction feeds signature detection, and validated documents feed automated upload.
Phase 01
Real-Time Fax Processing System
Integrated RingCentral webhooks so every incoming fax triggers processing immediately. Built an asyncio-based queue for concurrent handling, stored fax metadata in PostgreSQL for auditability, and archived all documents in AWS S3 with real-time status tracking at every stage.
Phase 02
AI-Powered OCR Extraction Pipeline
Deployed AWS Textract with SNS/SQS for scalable async OCR jobs. Built regex-based classification across 20+ document types with computer vision for signature detection. Every extraction outputs structured JSON with confidence scoring, fallback logic, and automatic retries on failure.
Phase 03
Outbound Fax Transmission
Built Playwright automation to retrieve pending orders directly from the healthcare platform, extract physician details, download PDFs, and dispatch via RingCentral or HumbleFax APIs. Includes dynamic provider selection, resend handling, and delivery confirmation tracking.
Phase 04
Automated EMR Upload Workflow
Playwright logs into the healthcare platform, searches by patient name, MRN, or order number, maps dynamically to the correct task type, and uploads the document with full error recovery and stealth operation mode — no API access required, no manual steps involved.
Phase 05
Centralized Monitoring and Optimization
Unified logging across all workflows with real-time dashboards, error categorization, Slack alerts, retry logic, and throttling. Every action from fax receipt to EMR upload is traceable, auditable, and compliance-ready from day one.
Have a Similar Use Case in Mind?
Let’s Build It.
Every automation we build starts with a single conversation. Tell us what your team is doing manually today,
and we will show you what an AI system could look like for your use case.
Tools & technologies
Tools and Technologies We Used
Every tool in this stack was chosen for a specific reason — reliability at healthcare scale,
HIPAA-aware data handling, and the ability to process documents without manual intervention.
What Changed, in the Client’s Words
The most impressive thing about Intuz is their ability to convert my vision into reality. I came in with a vague vision. The team was patient in getting that vision out of me and into the documents. Then, they turned it into reality. I used to have a five-person team whose job was to read and send out faxes. Now I have half a person assigned to that. We’re looking at $250,000 in savings per year — and that’s just one of the projects. The good part is I don’t have to scale up or down my team because the automation is running so efficiently. Even when I have to scale it, it’s going to be much cheaper to scale.
Intuz delivers on time and on budget. We’ve had delays, but those aren’t related to project management — it’s standard practice in the industry to try an approach and move on if it doesn’t work. That’s just how good engineering works.
