HealthcareClinicalRCMEDIHL7FHIRDjangoMongoDBRedisAI / speechHIPAA-aware

US Healthcare Platform — Clinical Operations & Revenue Cycle

Live multi-practice clinical and revenue-cycle platform: charting, nurse and patient dashboards, AI dictation, nurse/discharge AI summaries, and enterprise RCM on a practice-scoped Django/Mongo stack. Client identity and PHI omitted.

Prior senior engineering engagement — not an AEK-owned product. The platform is in production across multiple practices with substantial live clinical and billing data. AEK Tech’s co-founder contributed hands-on implementation across the stack while employed on that programme. Modules led end-to-end include patient survey, PI tracker, nurse dashboard, patient dashboard, AI dictation, clinical notes, and AI-generated nurse / checkout discharge summaries. Substantial ongoing work also covered RCM worklist, payment posting and adjacent clinical/billing flows. Presented here as capability proof for regulated health IT — with client identity and PHI omitted.

Status
In production
Region
United States
Timeline
Multi-year · ongoing
Team
Embedded engineering on client platform
  • Django 4
  • MongoDB
  • Redis
  • Amazon Transcribe
  • Deepgram
  • OpenAI SDK
  • X12 EDI
  • HL7
  • FHIR
  • S3
  • Snowflake ETL
  • Docker

US healthcare operators need one surface for multi-setting care and revenue cycle — not disconnected registration, orders and billing tools.

This case study documents senior engineering delivery on a practice-scoped clinical and RCM platform that is in production today: multiple practices on a shared multi-tenant model, real clinical and claims volume, shared member charting, order-set protocols, ED/inpatient/outpatient flows, and an X12 claim lifecycle with concurrent worklist controls. Clinical documentation includes AI dictation and speech-to-text (Amazon Transcribe and Deepgram), plus LLM-assisted nurse summaries and checkout / discharge summary generation. Integrations span HL7, FHIR chart pulls and clearinghouse-style eligibility and claims. The write-up stays deliberately generic — no client brand, no facility names, and no product screens — because this was employer delivery work, not an AEK product or public demo.

“In production across multiple practices — shared clinical and RCM workflows on one multi-tenant stack.”

“Speech AI with Amazon Transcribe and Deepgram — dictation into notes, plus nurse and discharge AI summaries.”

“X12 837P/837I generation with 835-driven payment posting and duplicate-safe worklist locks.”

The problem

Before consolidation, care settings, protocols and billing ran as fragmented workflows — duplicate work, reject-prone claims and weak remittance reconciliation when multiple billers touched the same encounter.

AreaProblemImpact
Multi-setting careSeparate ED / inpatient / outpatient flows without a shared member chartFragmented documentation, duplicated registration, weak handoffs
Orders & protocolsAd-hoc labs, imaging, meds and diagnoses per encounterInconsistent protocols; slow nursing and provider workflows
Claim creationManual CMS-1500 / UB-04 style entry with weak scrub/validationHigh reject risk; slow bill-to-pay cycle
RemittancePaper or manual 835 matching to open claimsDelayed payment posting; reconciliation errors
Concurrent billingMultiple billers editing the same encounter or claimDuplicate claims, overwrite races, unclear ownership
External systemsEHR charts, e-prescribe, clearinghouse and labs not on one ops surfaceSwivel-chair work; missed eligibility and documents
AnalyticsFacility A/R and census trapped in extracts and spreadsheetsWeak ATB visibility; slow operational decisions
Access controlBroad staff access without practice and role gatesLeast-privilege risk in a multi-practice deploy

Visuals

Product screens are intentionally omitted. This was confidential employer delivery on a live multi-practice system — not an AEK product. Ethics and agreements take priority over portfolio visuals. Module detail below stands in for screenshots.

Modules & capabilities

Member chart & registration

  • Master patient index / search; registration and coverage capture
  • Chart dashboard: diagnoses, documents, allergies, immunizations, letters, consents
  • Eligibility checks via clearinghouse-style APIs
  • Document storage via S3-compatible object storage (presigned access)
  • Wristband/barcode and facesheet-style print flows

Clinical dashboards, notes & AI (modules led)

  • Nurse dashboard and patient dashboard for day-to-day clinical ops
  • Clinical notes with structured documentation paths
  • AI dictation / ambient speech-to-text via Amazon Transcribe and Deepgram
  • LLM-assisted nurse AI summary generation
  • Checkout / discharge summary generation with AI assists
  • Structured extraction and ICD/SNOMED/CPT coding suggestion helpers
  • Patient survey flows for structured feedback collection
  • PI tracker for quality / performance improvement follow-up

Care settings (ED / inpatient / outpatient)

  • Tracking boards, triage, nursing assessments, disposition/discharge
  • Bed/room allocation and census history across settings
  • Specialty assessment packs on ambulatory and ED paths
  • Payment-link hooks from checkout-style flows

Order sets & ancillary services

  • Reusable order-set catalog (labs, imaging, meds, procedures, supplies, ICD)
  • Apply protocol to encounter; line-item status and completion
  • Laboratory worklist: specimen → transport → result → QC
  • Radiology worklist: order → schedule → report → study status
  • Pharmacy formulary/stock, med reconciliation, e-prescribe partner bridge

Scheduling & visit documentation

  • Provider calendars, slots, blocks, overbooks; appointment types and templates
  • Visit-linked meds, labs, vitals and SOAP
  • Ambient/dictation (Transcribe / Deepgram) → structured extraction
  • AI nurse summaries and checkout / discharge summary drafts
  • ICD/SNOMED/CPT suggestion helpers; visit → claim draft from encounter

HIM, worklist & prior auth

  • HIM master chart/patient index with completeness feedback
  • Assignable RCM/clinical worklist — document status, eligibility, claim create gates
  • FHIR/EHR patient and document retrieval for validation
  • Prior authorization / 278-style request tracking; referrals UI

Revenue cycle & claims (ongoing contribution)

  • Claim worklist: draft → scrub → validate → submit (837P and 837I)
  • ERA/835 ingest, remittance match, payment posting (draft → posted)
  • Charge master / fee schedule; patient price estimator
  • ATB-style facility dashboard; eligibility, denial and A/R style reports
  • Edit locks with heartbeat, reservations and duplicate-PCN guards

Integration gateway & ops

  • HL7 intake (ADT, SIU, ORM/ORU, MDM and related types) + X12 endpoints
  • Clearinghouse REST for claims, ERA, eligibility, enrollment/appeals
  • SFTP/S3 file pipelines for BI extracts and claim response attribution
  • Snowflake warehouse ETL with DEV → STG → PROD promotion
  • Audit logs, work queues and webhook-driven partner events

Technology choices and outcomes

ComponentTechnologyWhy it matters
ApplicationDjango 4, Gunicorn, WhiteNoiseMonolith web app for clinical ops and RCM in one deployable surface.
DataMongoDB (Djongo + PyMongo), RedisDocument models for charts/claims plus cache for session and high-read paths.
UIDjango templates, Bootstrap 4, jQueryServer-rendered clinical and billing worklists with AJAX data tables.
Authdjango-allauth, django-otp (TOTP / email OTP)Login, MFA and account lockout patterns for staff access.
JobsAPScheduler with Mongo leader leaseIn-process scheduled validation, reminders and file ingest without Celery.
EDI / interopX12 837P/837I/835, HL7 gateway, FHIR + OAuthClaim generation, remittance ingest and clinical messaging / EHR chart pull.
Cloud I/OS3-compatible storage, boto3, SFTP pipelinesClinical documents, EDI archives and BI extract attribution.
Speech AIAmazon Transcribe, DeepgramClinical dictation and ambient speech-to-text into charting workflows.
Clinical AI summariesOpenAI SDK (and related LLM paths)Nurse AI summary generation, checkout / discharge summaries, coding assists into structured notes.
WarehouseSnowflake Python ETL (DEV → STG → PROD)Cross-source analytics promotion without exposing warehouse branding here.

Security model

HIPAA-aware controls visible in the stack include practice-scoped RBAC, MFA options, audit logs, API-key gated interop and encryption tooling. No certification, BAA or partnership claims are made.

  • Practice/clinic-scoped multi-tenant queries on a shared database
  • URL-level RBAC plus fine-grained claim and worklist permissions
  • TOTP and email OTP MFA with account lockout after failed logins
  • Audit logs with action, module, old/new state, practice and actor
  • API-key and webhook-secured integration endpoints (not browser sessions)
  • Fernet and CSFLE toolkit present — not claimed as universal PHI encryption
  • HIPAA-aware engineering patterns only — no certification claims

Hardest technical challenges

The hard part was not drawing screens. It was making cross-framework compliance, workflow enforcement, and AI assists behave predictably in a regulated environment.

01

Concurrent claim creation without duplicates

Worklist reservations, edit locks with heartbeat, and practice-scoped PCN uniqueness guards address race conditions when multiple billers act on the same encounter. Idempotent create-claim paths reduce double-submit under slow clearinghouse responses.

02

Professional vs institutional EDI correctness

837P (CMS-1500 / SV1) and 837I (UB-04 / SV2) share infrastructure but diverge in loops, claim-level dates and institutional-only segments. Generation and archive jobs must keep form-type branching consistent with payer and clearinghouse expectations.

03

Remittance matching & payment posting lifecycle

835/ERA payloads must reconcile to open claims despite ID type mismatches and multi-ERA check numbers. Draft vs posted posting states and duplicate remittance skips protect A/R integrity.

04

Multi-tenant practice scoping on a shared MongoDB

Session-selected practice/clinic filters nearly every query; worklist includes explicit cross-practice access checks. Dual access (Djongo ORM + raw PyMongo) keeps auth models and high-volume collections workable at scale.

05

Integration reliability across HL7, FHIR and clearinghouse

Inbound clinical and billing messages are API-key gated and written to object storage; outbound claims poll for responses; EHR FHIR pulls documents for worklist validation. Failures surface as retryable jobs, listener logs and audit events rather than silent drops.

06

Monolith module surface area

Claims and patient dashboards are large view/helper surfaces with feature flags and legacy sync paths. Operational discipline — locks, queues and scheduled sweeps — compensates for the absence of a message-bus microservice split.

Results

Qualitative outcomes — no invented user or revenue metrics.

MetricBeforeAfter
Care documentationSiloed by settingShared member chart + order-set protocols
Claim submissionManual form-heavy entryScrub → validate → X12 submit pipeline
RemittanceManual ERA matchingAutomated ERA ingest + payment posting drafts
Biller concurrencyOverwrite / duplicate riskLocks, reservations, duplicate guards
EligibilityAd-hoc payer checksIn-workflow single/bulk eligibility
Clinical documentationManual note entry and siloed dictationSpeech AI (Transcribe / Deepgram) + structured notes
Nurse / discharge summariesHand-written or copy-paste summariesAI-assisted nurse and checkout / discharge summaries
Clinical ops surfacesFragmented nurse/patient viewsDedicated dashboards + clinical notes + AI dictation
External chartsSwivel-chair EHR accessFHIR document retrieval into worklist
Analytics warehouseFragmented source pullsScheduled ETL with STG→PROD promotion

Where the platform is now

The platform remains live with the client, serving multiple practices and holding substantial production data. Patterns here transfer to other regulated health IT environments — including UK care and health-tech programmes — without claiming NHS endorsement, HIPAA certification for AEK Tech, or ownership of the client system.

Confidential US healthcare engagement (prior employment). Presented by AEK Tech UK Ltd as capability proof — not an AEK product. No client branding or PHI.

Building something in this space?

If you need clinical workflows, speech AI in charting, RCM automation or regulated health IT engineering, we can talk through what a first phase should include — without expecting you to share PHI.