🩺
AI MedAgent
◆ Patent Pending
Platform Overview
← Platform Overview | Gateway Demo | ◑ Sandbox | ↗ SDK Docs | ▶ Arcade | ✉ Newsletter | ⌂ danielpettus.com ● Live Demo
Proof of Concept | danielpettus.com
AI That Acts. Not Just Alerts.
AI Agents in Acute Care
Autonomous Therapy. Continuous Learning.
Vendor-agnostic. The platform acute care has been waiting 25 years for.
A post-CABG patient in the ICU requires continuous titration of six high-risk infusions. Reactive AI agents watch vital signs and labs in real time, surfacing nurse-confirmed titration recommendations. Simultaneously, a cloud-based deep learning engine continuously analyzes the patient’s full clinical trajectory, integrating outcomes data, pharmacogenomics, and population models, to proactively optimize the therapy regimen. When the AI recommends a medication change, it automatically routes an updated order to pharmacy for approval, updates the automated dispensing cabinet, and alerts nursing, all before the nurse would have noticed the drift. The nurse remains in control. The AI never stops learning.
● 6 Reactive Agents
▲ Deep Learning Engine Active
IHE PCD · HL7 v2.6 · FHIR R4
Post-CABG ×3 · ICU Day 1
ICU Bed 4 | Post-CABG LIVE
LEVO 4mcg/mL 0.08 mcg/kg/min INFUSION HR:82 MAP:72 SpO2:97 AI ENGINE 6 AGENTS + DEEP LEARNING
ECG  ·  CONTINUOUS  ·  LEAD II
🧠
Deep Learning Optimization Engine
Proactive therapy adjustment: pharmacy, dispensing cabinet, and nursing notification
☁ Cloud  ·  Analyzing
🧠
AI Engine
Deep learning
outcomes model
Monitoring
📋
CPOE
Updated med order
AI recommendation
Pending
💊
Pharmacy
Pharmacist review
and approval
Pending
Dispensing Cabinet
Formulary updated
automatically
Pending
👩‍⚕️
Nursing
Pump auto-programmed
Nurse notified handheld
Pending
Infusion Monitor
Pump telemetry vs.
vitals trend, every 30 min
Pending
❤
Patient
Optimized therapy
best possible outcome
Goal
🏠
Discharge & Home
Wearables connected
Cardiologist linked
AI continues at home
Connected
Live Demo
Real Cloud AI
in Action
▶
🧠 Deep Learning Insight
Continuous analysis active: integrating hemodynamic trends, pharmacokinetic modeling, and population outcomes data for this patient profile.
Sequence paused at Dispensing Cabinet  —  please make a selection
⚠  AI MedAgent  ·  Diversion Watch
The Score Says Investigate.
But Is It Actually Diversion?
LLM clinical reasoning  ·  real-time patient context  ·  plain-English determination
Drug diversion costs hospitals $72 billion annually. Average detection time: 24 months. Over 80% of incidents go undetected entirely. Established dispensing companies have invested heavily in diversion analytics and their systems are genuinely sophisticated — pulling from dispensing cabinets, EMR data, and behavioral patterns across multiple signals. The output is a risk score. A score tells you a clinician looks anomalous compared to peers. It cannot tell you whether that anomaly is clinically justified.
ML Risk Score Output
“Clinician accessed fentanyl 3× above peer average this shift.”
Risk score: 87. Recommend investigation.
A pharmacist now manually reviews every transaction to determine whether this was legitimate patient care or actual diversion. That review takes hours.
vs
AI MedAgent — LLM Reasoning
“Pull at 02:14 for Patient A. Pain score 8/10 at 01:50. Physician order confirmed. Waste witnessed.”
Clinically justified. No anomaly. ✓
Patient clinical trajectory cross-referenced in real time. No manual review required. No false investigation opened.
The distinction is not what data is collected — it is what happens next.

Current systems aggregate cabinet and EMR data to find statistical outliers and surface them as risk scores. That still leaves a human to determine whether the outlier reflects genuine diversion or legitimate clinical activity. The false-positive burden is real and well-documented.

AI MedAgent takes the same connected data and passes it through an LLM informed by clinical literature, DEA scheduling data, ASHP guidelines, and AHRQ patient safety research. The LLM does not score the pattern against peers. It reads the patient’s clinical story and determines, in plain English, whether this specific transaction makes sense for this specific patient at this specific moment. The investigation either opens or closes immediately — no manual reconciliation, no hours in spreadsheets, no false positives consuming pharmacy resources.
▶  See Diversion Watch run live in the demo
ICU Bed 4  |  Post-CABG ×3  |  MR# 2026-4471
● AGENTS ACTIVE
--:--:--
MAP
72
mmHg ≥65
SBP/DBP
128/74
≤140 mmHg
Heart Rate
82
60-100 bpm
Resp Rate
14
br/min ≥10
SpO2
97
% ≥94
Temp
37.2
°C
Glucose
118
80-140 mg/dL
RASS
-1
Target -1 to 0
Vasoactive / Cardiovascular
Norepinephrine
Levophed
Vasopressor
0.08
mcg/kg/min
MAP
72 mmHg
≥65
Watching: MAP within target
Nicardipine
Cardene
Vasodilator
2.5
mg/hr
SBP
128 mmHg
≤140
Watching: SBP within target
Esmolol
Brevibloc
Rate Control
50
mcg/kg/min
Heart Rate
82 bpm
60-100
Watching: HR within target
Sedation & Analgesia
Metabolic
Propofol
Diprivan
Sedation
20
mcg/kg/min
RASS
-1
-1 to 0
Watching: RASS at target
Fentanyl
Sublimaze
Analgesia
25
mcg/hr
Resp Rate
14 br/min
≥10
Watching: RR safe
Insulin Infusion
Regular Insulin
Glycemic
2.0
units/hr
Glucose
118 mg/dL
80-140
Watching: glucose in range
🔗 AI Connectivity Engine
IHE PCD-01
HL7 v2.6 pump telemetry
HL7 v2.6PCD TF-2
ICD-10 / SNOMED
Diagnosis context binding
Z95.1CABG
FHIR R4
MedAdmin & Observation
R4 RESTSMART
Deep Learning API
Cloud outcomes model
CLOUDASYNC
ai_connectivity_engine.py | IHE/HL7/FHIR + Deep Learning
■ AI GENERATING
The Full Closed Loop — From Order Entry to Bedside Infusion Monitor
One Order. Nine Steps.
AI at Every Stage.
This is a scripted demonstration, not a live system. It illustrates one full closed-loop sequence as the patented method is designed to work: a physician enters a post-CABG order set, specialized agents evaluate it against evidence, and a recommendation returns for physician review, pharmacy verification, and dispensing cabinet routing, ending with an infusion monitor agent checking pump telemetry against the vitals trend. Every response shown is fixed, pre-written example content, currently limited to this local demo. Live AI reasoning has been disabled here due to cost and maintenance constraints for a solo-run project. That scripted sequence is one click away.
Physician + AI
Order entered in CPOE. Four AI agents query CMS, AHRQ, NIH, and FDA in parallel. Evidence-based recommendation returned in seconds for one-click physician review.
Pharmacy + AI
A second AI agent verifies dilution, checks renal dose for CrCl 60-75, and scans all six active infusions for interactions before the pharmacist approves.
Dispense + Notify
Dispensing cabinet unlocked automatically. Nurse alerted via PCD-06. Smart pump library queued. No phone calls. No manual transcription. No delay.
Every Decision Trains the Model
Accept, override, or modify — every outcome is recorded. The system improves with each patient encounter. Lane keeping today. Full self-driving tomorrow.
Infusion Monitor + AI
Thirty minutes post-dispense, a third AI agent cross-references smart pump telemetry against the vitals trend and recommends maintain, titrate up, or titrate down, with a one-click nurse confirmation or manual override.
⚕
Illustrative Walkthrough  ·  Patented Method
See the Full Loop, As Designed
Physician order to pharmacy verification to dispensing cabinet — eight steps, illustrating the reasoning method described in the patent filing. Not connected to live hospital data, external databases, or cloud infrastructure.
▶
Launch Demo
Resources & Partnership

Review the Work. Start a Conversation.

The platform specification, SDK, and architecture are available for review. These are development-stage documents representing a concept that is ready for the right partner to take to production.

Development-Stage Documents
🎭
Strategic Partnership Presentation
The clinical problem, the inflection point, the CABG example, platform architecture with component diagram, where we are today, and the partnership opportunity.
8 SLIDES DEVELOPMENT STAGE PDF ↓
Development Stage Disclosure
These documents represent a concept platform in development. Architecture, specifications, and interfaces are illustrative. No production software or infrastructure currently exists. All IP, architecture, and designs are the property of Daniel C. Pettus.
General Inquiry