Nurse Amy – A Hallucination-Free, HIPAA-Compliant AI Agent

Overview

Our client, one of the largest RPM providers in the country, came to us with a very specific problem: they were responsible for communicating with thousands of patients across the country, and they needed to do it safely, accurately, and at scale. Ensuring patient satisfaction was their top priority, but so was maintaining HIPAA compliance. They needed help managing everything from daily patient reminders to structured data collection—and they needed to eliminate the risk of hallucination, privacy violations, and inconsistency in communication.

I built Nurse Amy to solve exactly that.

Nurse Amy is a HIPAA-compliant, ontology-driven AI agent designed to perform real patient engagement tasks, not just carry on conversations. She automates repetitive, regulated workflows while remaining fully auditable and hallucination-free. She now operates at Medek Health Systems, handling patient communication and logging for RPM, RTM, and CCM workflows. Her work supports over a million claims, improving operational efficiency and documentation integrity.

‍

The Problem

There were three main issues our client faced:

• Routine patient interactions were manual and time-intensive: Reminding patients to take readings, helping with devices, and checking symptom compliance all required human effort.
• Existing AI tools were unreliable or insecure: Anything based on generative AI posed a real liability. These models could invent information or inadvertently offer medical advice—a nonstarter under HIPAA.
• Accurate documentation was critical for billing: Every interaction needed to be logged in structured form to support CPT codes like 99490, 99453, and 99457.

Off-the-shelf solutions fell short. None were HIPAA-compliant, and none could be integrated safely with protected health data. Worse, they couldn’t be trusted to execute deterministic logic reliably—let alone log it all in an EHR.

‍

What I Built

Nurse Amy is a domain-specific AI agent that understands natural language, performs real actions, and logs everything with full traceability. She’s not just a voice assistant. She’s a structured interface between patients and the system.She handles:

• Patient reminders: Contacts patients to remind them to take RPM readings or attend CCM appointments, with logic tailored to the care plan.
• Device troubleshooting: Guides patients through workflows for glucometers, BP cuffs, or other RPM hardware, escalating only when necessary.
• Data extraction and transcription: Transcribes patient calls, extracts relevant information, and logs structured notes directly into the EMR.
• Symptom surveys: Conducts brief, compliant check-ins and surveys to satisfy billing documentation requirements.
• Smart escalation: Knows when she’s out of depth, and routes the patient to a human operator with full conversation history and context.

‍

Why It Works

Nurse Amy runs on a deterministic stack designed by Intelligence Factory:

• SemDB provides semantic retrieval and billing-aware context
• OGAR powers intent classification and conversation flow based on medical ontologies
• Buffaly handles secure logic execution and structured integration with EHR systems

There are no calls to OpenAI. No black-box models. Every decision Nurse Amy makes is traceable to a rule, a script, or a structured care plan.

‍

Results at Medek

After full deployment, Nurse Amy now fills six patient-facing roles at Medek:

• Reminder scheduling and outreach
• RPM device troubleshooting
• Patient surveys
• Real-time transcription and note logging
• Patient queries and FAQs
• Controlled escalation for out-of-scope scenarios

The impact:

• Measurable higher patient satisfaction.
• Elimination of customer service / patient interaction adverse “incidents”
• Support for over 1M claims
• Improved claim success rate with fewer documentation errors

‍

Why We Didn’t Use Existing Tools

Because you can’t plug a HIPAA-regulated EHR into ChatGPT. You can’t let an LLM offer medical advice. And you can’t rely on a black box to meet insurance audit standards.I didn’t build Nurse Amy to be clever. I built her to be correct. She’s grounded, deterministic, and capable of acting on and recording patient interactions safely and effectively.

‍

The Takeaway

If you want a general-purpose bot, you can buy one. But if you need a safe, accurate, audit-ready system for regulated healthcare workflows, you need to build it like we did—from first principles, with ontologies, local execution, and strict control.That’s Nurse Amy. And she’s just the beginning of what Perfect Agents can do.