Healthcare Engineering
We Build AI-Powered Clinical Platforms for PACE Organizations. Not Generic Healthcare Software.
The Problem
Your CRS team is toggling between myUnity, Excel trackers, and email chains to chase HCC codes. Your enrollment pipeline crosses 11+ disconnected systems — some undocumented, maintained by a single developer. Monthly CMS submissions involve manual reconciliation of MODR, MMR, and MONMEMD files. Your IDT meetings produce action items that get lost between systems before the next huddle.
You've talked to vendors. The EHR companies treat PACE as an add-on module. The CRM companies track your enrollment funnel but don't understand nursing-home level of care determinations. The AI vendors pitch flashy demos but can't explain how their model handles PACE-specific HCC hierarchy suppression or why provider attestation must remain human-verified under 42 CFR 422.310.
Closing the Revenue Gap in PACE Organizations
Missing a single HCC per participant costs $500–$2,000/month in lost capitation. For a 500-participant organization with 4 missed HCCs on average, that's $1M–$4M/year left on the table. The platform gap isn't just operational — it's a revenue problem that AI can directly address.
Who We Are
AI for PACE Healthcare
AI That Understands PACE Operations — Not AI That Learned Healthcare from a Textbook
Every AI solution below was designed while building production PACE workflows. We didn't start with a model and look for a use case — we started with your pain points and designed AI that fits inside the clinical workflow, the compliance boundary, and the revenue model. Here's what that looks like:
The Pain
HCC Codes Missed at Every Encounter
Providers document in free text. CRS reviews after the fact. By the time someone notices a missed CHF or diabetes diagnosis, the encounter is closed, the superbill is submitted, and $500–$2,000/month per participant is lost from your capitation payment.
What We Built
Real-Time HCC Capture Coaching
AI monitors documentation during the encounter — not after. When a provider mentions CHF symptoms, the system prompts: "Document assessment to support HCC 85 — this adds $300/month to capitation." Live RAF preview updates as diagnoses are added, showing V22 and V28 impact side by side.
The Pain
Clinical Documentation Takes Too Long
Providers spend 30–40% of their day typing notes. SOAP-format documentation is repetitive. Copy-paste from previous visits leads to stale information and compliance risk. Meanwhile, 10 participants are waiting in the day center.
What We Built
AI-Powered Documentation Assistant
Voice-to-text transcription with smart clinical parsing — provider speaks naturally, AI structures into SOAP format. Dynamic templates auto-populate from the participant's problem list, recent labs, and upcoming due dates. "This participant needs flu shot (due), diabetes foot check (overdue), PHQ-9 (due)."
The Pain
Enrollment Loses Revenue from Day One
When a new participant enrolls, the physician manually reviews referral records and documents diagnoses. HCCs that exist in the referral paperwork never make it to the initial encounter. You're underpaid from the first capitation check.
What We Built
AI-Assisted HCC Capture at Enrollment
NLP extracts diagnoses from referral records, hospital discharge summaries, and prior authorization documents. Presents candidates to the physician for confirmation — not auto-coding. "Referral documents mention COPD (J44.1) and Type 2 Diabetes (E11.9) — confirm to capture HCCs 111 and 19."
The Pain
External Encounter Reconciliation Is Manual
When a participant is discharged from a hospital, someone has to manually read the discharge summary, key in diagnosis changes, flag medication changes, and update the care plan. It takes hours per participant and delays IDT follow-up.
What We Built
NLP Discharge Summary Auto-Reconciliation
AI extracts key information from discharge summaries — new diagnoses highlighted, medication changes flagged for reconciliation, follow-up recommendations parsed and routed to the appropriate IDT member. Provider reviews and confirms, but 80% of the data entry is eliminated.
The Pain
ICD-10 Coding Is Inconsistent and Slow
CRS reviews encounters and manually maps diagnoses to ICD-10 codes, then checks HCC crosswalks. Different CRS staff code the same condition differently. MEAT criteria flagging is subjective. The whole superbill pipeline bottlenecks at coding review.
What We Built
AI/NLP ICD-10 Code Extraction & Suggestion
NLP extracts ICD-10 candidates directly from clinical notes. AI suggests the most specific code with HCC mapping preview — CRS reviews, accepts, or overrides. Structured MEAT criteria prompts ensure documentation supports every diagnosis. Every suggestion is auditable; no code reaches a claim without human validation.
The Pain
Multilingual Communication Gaps
PACE serves diverse elderly populations. Pharmacy labels, care plan summaries, and family portal content need to be in the participant's language. Today it's either English-only or manual translation by bilingual staff — inconsistent, slow, and risky for medication safety.
What We Built
AI-Assisted Multilingual Document Engine
Three-tier translation architecture: curated medical library first (verified clinical terminology), AI-assisted translation second (for dynamic content), with human validation required before any translated content reaches a participant. Serves 10+ use cases beyond pharmacy labels — care plan summaries, appointment reminders, family portal content, consent forms.
AI for PACE Healthcare
AI That Understands PACE Operations — Not AI That Learned Healthcare from a Textbook
We keep it simple and move fast.
Step 01
15-min Domain Check
You describe your PACE operation — systems, pain points, AI readiness. We ask questions a generic agency wouldn't know to ask.
Step 02
Architecture Alignment
We walk through team structure, integration points, security requirements, and AI opportunities. You'll see we've been inside this problem before.
How We Keep AI Safe in Clinical Environments
Human-in-the-Loop — Always
AI suggests, clinician confirms. NLP-extracted ICD codes are candidates until a CRS accepts them. Translated pharmacy labels require human validation before printing. No exceptions.
Auditable AI Decisions
Every AI suggestion is logged — what was suggested, what was accepted, what was overridden, by whom, with timestamp. Immutable audit trail meets 42 CFR 460 and CMS RADV requirements. AI decisions are replayable for compliance review.
PHI Never Leaves the Boundary
AI models operate within the platform's security perimeter. No PHI is sent to external AI services without explicit consent and BAA coverage. KMS encryption, tenant isolation, and DLP hooks apply to AI pipelines identically to clinical data flows.
Security & Compliance
Security Isn't a Feature We Add. It's the Foundation Everything Sits On.
In PACE, a misconfigured access boundary doesn't create a support ticket — it creates a HIPAA breach affecting a vulnerable elderly population. Every architectural decision we make starts with the compliance question, not ends with it.
PHI Encryption — Everywhere
AES-256 encryption at rest via AWS KMS. TLS 1.2+ for all data in transit. Per-tenant encryption keys in Secrets Manager. No plaintext PHI exists anywhere in the stack — not in logs, not in caches, not in error messages.
Tenant Data Isolation
Per-tenant PostgreSQL databases — not row-level filtering. Customer A cannot see Customer B's data at any layer. Separate Cognito App Clients per tenant. Lambda execution context carries tenant_id from JWT — no cross-tenant leakage is architecturally possible.
Immutable Audit Trails
Every PHI access logged — who, what, when, why. Audit records are append-only and cannot be modified by any role. Clinical audit logs (CREATE, UPDATE, DELETE, SIGN, COUNTERSIGN, AMEND) retained for 10 years per 42 CFR 460. Point-in-time snapshots for CMS RADV audit replay.
Clinical RBAC
Role-based access aligned to real PACE org structures — CRS, providers, IDT coordinators, enrollment specialists, family portal users. Not generic "admin/user/viewer." Minimum necessary access enforced at API Gateway. MFA for workforce; simplified auth for family portal (HIPAA-appropriate).
Breach Detection & Response
CloudTrail for every API call. CloudWatch anomaly detection for unusual PHI access patterns. VPC Flow Logs for network-level monitoring. Automated breach notification workflows aligned to HIPAA timelines (60-day individual, 60-day HHS for 500+ affected).
Compliance-Ready from Day One
42 CFR Part 460 retention requirements built into data models. HIPAA Security Rule controls mapped to AWS services. BAA-eligible AWS services only. Enterprise security review documentation pre-built. Your compliance team won't find gaps — because we built to their checklist, not ours.
Domain Proof
We Haven't Just Read About PACE. We've Architected for It.
Every proof point below comes from active engineering work — production specs, validated data models, and real clinical workflow mapping. Not slide decks.
The Revenue Case for AI-Powered Platform Modernization
PACE organizations receive capitated payments scaled to each participant's RAF score. AI closes the gap between documented diagnoses and submitted HCCs at every touchpoint — enrollment NLP capture, real-time encounter coaching, automated ICD suggestion, and CMS reconciliation. The ROI isn't theoretical; it's per-participant-per-month math.
95%
Target HCC Capture Rate
100%
Attested Submissions
$1M–$4M
Annual Revenue at Risk (500 ppts)
