PULSE_AGENTS

Pulse Agents

19 AI agents deployed on the REST gateway (/linc/{agent}) delivering FHIR R4 clinical decision support — including 14 clinical agents, 5 predictive models, HuggingFace model bridge, and conversational chat.

Agent Overview

Clinical Agents (12)

#	Agent	Endpoint	FHIR Resources	Clinical Domain
1	Summary	/linc/summary	DocumentReference, Composition	Patient summary tailored to role (doctor, patient, care-manager)
2	Prior Auth	/linc/prior-auth	Claim, ClaimResponse	Prior authorization copilot with coverage checks
3	Gaps in Care	/linc/gaps-in-care	CarePlan, Observation	Preventive screening gap finder
4	Medication Safety	/linc/medication-safety	MedicationRequest, AllergyIntolerance	Drug-drug interaction and allergy checks
5	Care Plan	/linc/care-plan	CarePlan, Goal	AI-powered care plan navigator
6	Clinical Trials	/linc/clinical-trials	ResearchStudy, Group	FHIR clinical trial matcher
7	Readmission Risk	/linc/readmission-risk	RiskAssessment	Hospital readmission risk workbench
8	Triage	/linc/triage	Encounter, Observation	Conversational triage assistant
9	Imaging Followup	/linc/imaging-followup	ImagingStudy, DiagnosticReport	Imaging results follow-up tracker
10	Lab Explainer	/linc/lab-explainer	Observation, DiagnosticReport	Patient-friendly lab result explanations
11	NL Query	/linc/nl-query	Parameters	Natural language to FHIR query explorer
12	SDOH Referral	/linc/sdoh-referral	ServiceRequest, Task	Social determinants and community referral matcher

Predictive Models (5) — NEW

#	Agent	Endpoint	FHIR Resource	Algorithm	Key Output
13	Predict Readmission	/linc/predict-readmission	RiskAssessment	Patient ID hash + age + comorbidity count	30-day risk score + mitigations
14	Predict PA Denial	/linc/predict-pa-denial	RiskAssessment	Service code + payer + patient history	Pre-submission denial probability
15	Predict ED Util	/linc/predict-ed-util	RiskAssessment	Age + seasonality + prior ED visits	7d and 30d ED visit forecast
16	Predict Interaction	/linc/predict-interaction	RiskAssessment	Warfarin+NSAID+ACE+diuretic rules	4 drug-drug interactions with severity
17	Predict No-Show	/linc/predict-no-show	RiskAssessment	Prior no-show + lead time + transport access	No-show probability + interventions

Integration Agents (2)

#	Agent	Endpoint	FHIR Resource	Description
18	Chat	/hf/chat	Parameters	Conversational AI with clinical reasoning
19	HF Models	/hf/models	Parameters	HuggingFace model bridge status

Usage

# List all agents
curl http://localhost:58080/linc/ | jq .

# Patient summary (doctor role)
curl "http://localhost:58080/linc/summary?patient=P-5842&role=doctor" | jq .

# Predictive: Readmission risk (30-day)
curl "http://localhost:58080/linc/predict-readmission?patientId=P001" | jq .

# Predictive: Drug interaction check (4 interactions)
curl "http://localhost:58080/linc/predict-interaction?patientId=P005&medications=metformin,lisinopril,warfarin,atorvastatin" | jq .

# Predictive: No-show probability
curl "http://localhost:58080/linc/predict-no-show?patientId=P003&appointmentDate=2026-06-15" | jq .

# Predictive: PA denial probability
curl "http://localhost:58080/linc/predict-pa-denial?patientId=P004&service=CT%20Abdomen&payer=Medicare" | jq .

# Predictive: ED utilization forecast
curl "http://localhost:58080/linc/predict-ed-util?patientId=P002" | jq .

# Chat with AI
curl "http://localhost:58080/hf/chat?q=what%20is%20normal%20blood%20sugar" | jq .

# Check HF models status
curl http://localhost:58080/hf/models | jq .

Batch Test All 19 Agents

for agent in summary prior-auth gaps-in-care medication-safety care-plan clinical-trials readmission-risk triage imaging-followup lab-explainer nl-query sdoh-referral chat hf-models predict-readmission predict-pa-denial predict-ed-util predict-interaction predict-no-show; do
  STATUS=$(curl -s -o /dev/null -w "%{http_code}" "http://localhost:58080/linc/$agent?patientId=P001" 2>/dev/null || echo "FAIL")
  echo "$agent → HTTP $STATUS"
done

Predictive Model Details

Readmission Predictor

{
  "resourceType": "RiskAssessment",
  "prediction": [{ "probabilityDecimal": 0.95, "when": { "unit": "d", "value": 30 } }],
  "riskFactors": ["Age > 65", "Prior admission within 90d", "3+ comorbidities"],
  "mitigations": ["Daily check-in call", "7-day post-discharge follow-up"],
  "model": "readmission-v2.1",
  "confidence": "moderate"
}

Drug Interaction Predictor

{
  "resourceType": "RiskAssessment",
  "predictions": [
    { "medications": "warfarin+NSAID", "risk": "high", "effect": "Increased bleeding risk" },
    { "medications": "lisinopril+NSAID", "risk": "moderate", "effect": "Reduced antihypertensive effect" },
    { "medications": "metformin+contrast", "risk": "moderate", "effect": "Lactic acidosis risk" },
    { "medications": "lisinopril+potassium", "risk": "low", "effect": "Hyperkalemia risk" }
  ],
  "model": "interaction-v1.0"
}

Agent Architecture

Each agent:

1. Receives a FHIR R4-style request with patient ID and optional parameters
2. Attempts to fetch live data from https://hnh.brainsait.org/api endpoints
3. Falls back to clinically-accurate synthetic data when live data is unavailable
4. Returns a standard FHIR resource response with live_data: true|false indicator
5. All responses include resourceType and FHIR-compliant structure

Predictive models use deterministic algorithms (patient ID hash + age) rather than requiring ML infrastructure.

HuggingFace Integration

Three HuggingFace model repositories are integrated:

Model	Repository	Task
Meditron	Fadil369/meditron-clinical-guidelines-bucket	Clinical guidelines for Saudi healthcare
LLaVA-Med	Fadil369/llava-med-v1.5-mistral-7b-bucket	Medical imaging visual QA
Medical QA	Fadil369/medical-question-answering-datasets	Patient inquiry resolution

Access: http://localhost:58080/hf/models (requires HF_TOKEN for private repos)

Chat AI Capabilities

The chat agent (/hf/chat) understands:

• Clinical triage — Chest pain, cardiac symptoms, emergency protocols
• Lab interpretation — CBC, BMP, LFT, Lipid panel, cardiac biomarkers
• Medication safety — Drug interactions, dose adjustments, Saudi FDA alternatives
• Appointments — Scheduling and availability checks
• Claims — NPHIES claim lifecycle and status tracking
• General health queries — Education and guidance

Response Format

All agents return FHIR R4 Parameters or domain-specific resources:

{
  "resourceType": "Parameters",
  "parameter": [
    { "name": "response", "valueString": "..." },
    { "name": "patient", "valueString": "P-5842" },
    { "name": "live_data", "valueBoolean": false }
  ]
}

Predictive models return RiskAssessment:

{
  "resourceType": "RiskAssessment",
  "status": "final",
  "subject": { "reference": "Patient/P-5842" },
  "prediction": [{ "probabilityDecimal": 0.85, "when": { "unit": "d", "value": 30 } }],
  "model": "model-v1.0",
  "confidence": "moderate"
}