Patients Today
47
32 routine · 8 emergency · 7 new patients
Radiograph Findings
3
AI-detected pathology requiring review
UDA Completion
847
This month · 94% of contract target
Recall Due
284
Overdue recall appointments
🤖 AI Agent Status — Today
9 agents covering clinical intelligence, NHS contract, and practice operations
Radiograph AI Agent3 findings flagged · Dentist review
Clinical Notes Agent41 notes drafted · 89% accepted
NHS Contract IntelligenceUDA tracking live · 94% on track
Periodontal AI Agent8 BPE codes updated
Patient Recall Agent47 recall SMS sent today
Treatment Planning Agent14 treatment plans drafted
📡 Live Clinical Feed
Real-time AI agent actions — all clinical outputs require dentist review
Why DentistOS — The Dental Practice AI Opportunity
🩻 Radiograph AI
Dentists review 8-12 radiographs per session under time pressure. Studies show 24% of pathology is missed on bitewing radiographs during busy sessions. DentistOS Radiograph AI detects caries (94% sensitivity), periapical pathology, bone loss patterns, and root fractures — presented as a second opinion, never overruling the dentist.
💷 NHS UDA Contract Leakage
The average NHS dental practice leaves 8-12% of contracted UDAs undelivered due to poor appointment scheduling and patient did-not-attend (DNA) rates. NHS Contract AI maximises UDA delivery rate, predicts DNA patients for proactive confirmation, and flags underperformance against contract 8 weeks before financial year-end — when it can still be recovered.
📞 Patient Recall Failure
The average dental practice has 28% of its patient base overdue for recall. Each missed recall is a missed appointment slot, missed clinical opportunity, and missed revenue. Patient Recall AI segments patients by risk (medical complexity, last attendance, systemic health status) and sends intelligent, personalised recall messages — achieving 67% higher response rate than generic recall letters.
DentistOS AI Agents — Dental Intelligence Layer
Radiograph AI Agent
BW, PA, and OPG radiograph analysis. Caries detection (proximal, cervical, occult) — 94.1% sensitivity vs 78% human baseline under clinical conditions. Periapical pathology detection. Alveolar bone level measurement. Root fracture probability. Implant bone interface. All findings flagged for dentist review — AI never makes clinical diagnosis.
Running · 3 findings today
ReAct + RAGPeriodontal AI Agent
Six-point pocket chart AI interpretation. BPE code assignment and tracking. BSP 2019 periodontal classification (Stage/Grade). Treatment pathway recommendation (S1-S4). Furcation involvement mapping. Bone loss rate calculation. Periodontal risk assessment (Tonetti model). NICE NG30 periodontal disease guideline integration.
Running · 8 BPE updated
Reflection + RAGClinical Notes Agent
Dental consultation → structured clinical note (charting, periodontal data, treatment provided, advice given, next appointment). Dental notation (FDI/Palmer). FP17/FP17PR form auto-completion. NHS/private fee posting. Consent documentation. Medical history alert integration. Average note time: 6 min → 45 seconds.
Running · 41 notes today
ReAct + RAGTreatment Planning Agent
Comprehensive treatment plan generation: diagnosis, treatment options (NHS vs private), cost estimate, sequencing, and expected outcomes. Phased treatment planning for complex cases. Restorative risk analysis. Implant suitability assessment. Written patient information letters. Estimates and consent forms auto-generated for dentist review and signature.
Running · 14 plans drafted
Planning + RAGNHS Contract Intelligence Agent
Real-time UDA delivery tracking vs contract. Band distribution analysis (Band 1/2/3 mix). DNA prediction and appointment confirmation outreach. Financial year-end projection with recovery recommendations. Contract underperformance alert 8 weeks before year-end. Protocolised patient re-registration for open NHS access. Prevents NHS clawback penalties.
Running · 847 UDAs · 94%
Planning + Tool UsePatient Recall Agent
Risk-stratified patient recall: medical complexity (diabetes, anticoagulation, immunosuppression), treatment history, caries risk assessment (Cariogram), and periodontal risk. Personalised recall interval (3-24 months per NICE NG30). SMS + email + letter channels. DNA prediction. 67% higher recall response rate vs generic recall. Integrates with Dentally, Software of Excellence, and R4.
Running · 47 recalls sent
Planning + Tool UseMedical History Alert Agent
Automated medical history flag review at each appointment. Drug interactions (anticoagulants, bisphosphonates, SSRIs). Antibiotic prophylaxis requirements (NICE NG195). MHRA drug alerts. Allergy cross-reference (latex, materials, LA agents). Medically compromised patient protocol activation. Bisphosphonate MRONJ risk scoring.
Running · 47 patients checked
ReAct + Tool UsePractice Analytics Agent
Revenue per surgery, revenue per hour, new patient acquisition rate, recall response rate, failed appointment cost, treatment plan acceptance rate, associate vs principal performance, and patient satisfaction (Friends & Family Test). CQC inspection readiness score. Monthly practice health dashboard for principal dentist.
Idle · Weekly report
Planning + Tool UseCQC Compliance Agent
CQC Key Lines of Enquiry (KLOE) compliance tracking: Safe, Effective, Caring, Responsive, Well-led. Decontamination protocol compliance (HTM 01-05). Staff DBS and qualification currency. Complaint handling (within 20 working days). Incident reporting. GDC Standards for the Dental Team alignment. Inspection readiness score maintained real-time.
Idle · Inspection ready
RAG + Tool UseRadiograph AI — Clinical Decision Support
⚠ ALL RADIOGRAPH AI FINDINGS ARE CLINICAL DECISION SUPPORT ONLY. DIAGNOSIS IS ALWAYS MADE BY THE REGISTERED DENTIST OR SPECIALIST. AI OUTPUT REQUIRES CLINICAL REVIEW BEFORE PATIENT COMMUNICATION.
Radiographs Analysed
47
Today · BW + PA + OPG
Pathology Flagged
3
Requiring dentist review
Caries Detection Rate
94.1%
Sensitivity (validated dataset)
False Positive Rate
8.4%
All reviewed by clinician
🩻 Today's Flagged Findings — Dentist Review Required
Patient PT-2847 — UR7 Periapical Pathology
Periapical PA radiograph: 3mm periapical radiolucency UR7. AI confidence 0.87. Differential: chronic periapical abscess vs granuloma vs cyst. Clinical correlation recommended — pulp vitality testing. Root canal therapy vs extraction decision for dentist. Patient asymptotic.
Patient PT-1093 — LR5 Proximal Caries
Bitewing radiograph: mesial surface LR5 — radiolucency reaching outer third of dentine. AI confidence 0.79. ICDAS/ICCMS: D3 lesion likely. Clinical probing and transillumination recommended to confirm. Composite restoration indicated if confirmed.
Patient PT-3421 — Generalised Bone Loss
OPG: Generalised horizontal bone loss 20-30% across posterior sextants. Furcation involvement probable LL6 and UL6. BPE recorded as 3 today — may underestimate full extent. Full periodontal assessment (6-point chart) recommended. BSP Stage III Grade B classification likely.
📊 Radiograph AI Performance — Validated Metrics
Caries detection sensitivity94.1% (vs 78% human baseline)
Periapical pathology sensitivity89.3%
Bone loss quantification accuracy±0.8mm (vs ±2.1mm manual)
Root fracture probabilityProbabilistic score only
Implant bone interfaceCrestal bone level ±0.3mm
MHRA classificationDecision support — not SaMD
📋 GDC Standard 4.1: The registered dentist is responsible for the radiograph report. AI findings are flagged as clinical decision support. Every AI finding must be reviewed, accepted, modified, or rejected by the dentist before charting.
NHS Contract Intelligence — UDA Tracking & Optimisation
Annual UDA Contract
9,000
NHSE commissioned · FY 2025-26
UDAs Delivered YTD
5,847
Month 8 · On track ✓
Delivery Rate
94%
Target: 96% minimum
Clawback Risk
£12,400 / ₹13L / $15,750
If current DNA rate continues
DNA Rate
8.4%
National avg 11.2% — managing well
📈 UDA Delivery — Monthly Tracking
AI projects year-end performance and flags recovery actions
Month 8 actuals (delivered)847 UDAs · On track
Band 1 (1 UDA) — examinations312 (37%)
Band 2 (3 UDA) — fillings/extractions389 (46%)
Band 3 (12 UDA) — dentures/crowns146 (17%)
Months 9-12 projected UDAs3,247 (AI forecast)
Year-end total projected9,094 (101.0%)
💡 AI Recommendation: DNA rate of 8.4% is projected to cost 312 UDAs by year-end (£14,352 / ₹15.1L / $18,230 clawback risk). Sending confirmation SMS 48h before appointments for high-DNA-risk patients (identified by AI) has reduced DNA rate by 34% in trials.
📊 NHS Performance Dashboard
Weighted average UDA value£28.40 / $36 / ₹2,985 per UDA
Contract value (annual)£255,600 / $324,750 / ₹2.69Cr
Year-to-date income earned£165,941 / $210,750 / ₹1.75Cr
Year-to-date income at risk (DNA)£12,400 / ₹13L / $15,750
New patient registration (NHS)47 this month
Exempt patients (under 18, benefits)34% of list
CQC compliance score94/100 · Inspection ready
GDC continued registrationAll associates current ✓
AgentOps — Live Agent Observability
📡 Live Trace Feed
📊 Session Metrics (24h)
Total Sessions847
Avg Latency1.4s
P95 Latency3.1s
Error Rate0.3%
HITL Escalations12
RAGAS GatePASS ✓
💰 Cost & Tokens
Cost (24h)£124 / ₹13,000 / $155
Input Tokens8.2M
Output Tokens2.1M
Cache Hit Rate71%
Cost/Session£0.15 / ₹16 / $0.19
🎯 RAGAS Quality Scores
Faithfulness0.94 ✓
Answer Relevance0.91 ✓
Context Precision0.89 ✓
Hallucination Rate0.8%
Context Recall0.93 ✓
🤖 Agent Health
All agentsHealthy
OrchestratorActive
Tool registryOnline
MCP serversConnected
Memory storeHealthy
Guardrails — Responsible AI & Patient Safety
🛡 Active Safety Rails
✅ Clinician-in-the-Loop (HITL)
Every clinical AI recommendation requires authorised clinician review and sign-off before any patient-facing action. AI cannot autonomously update patient records, prescribe, or refer. GDPR Article 22 — no fully automated consequential healthcare decisions.
🔍 Patient Data — NHS DSPT / HIPAA
All patient identifiers (NHS number, name, DOB, address) stripped via Presidio before LLM processing. Data never leaves NHS-approved environment. No cross-patient data in AI context windows. DSPT Standards Met compliance.
⚠️ Clinical Safety — DCB 0129
All AI clinical recommendations carry confidence score + uncertainty disclosure. AI never overrules treating clinician. High-risk flags (suicide risk, safeguarding, emergency) automatically escalated to senior clinician within 60 seconds.
📝 Immutable Clinical Audit Trail
Every AI interaction logged with: patient pseudonym, input context, AI output, clinician response, and outcome. 10-year retention. Exportable for CQC, GMC, GDC, or NMC investigations.
📋 Regulatory Compliance Mapping
GDPR Art. 22 — Automated decisionsHITL enforced
NHS DSPT — Data SecurityCompliant
MHRA SaMD — Advisory onlyExempt
DCB 0129 — Clinical riskDocumented
EU AI Act — High-risk AIMapped
NIST AI RMFAligned
ISO/IEC 42001Compliant
0.8%
Hallucination Rate
Target <2%
100%
HITL Coverage
0
PII Leaks (30d)
A+
Security Grade
Documentation — Deployment Guide & Runbook
🚀 10-Week Deployment Guide
1
Week 1–2: Data Foundation & Integration
Connect to clinical systems (EMIS/SystmOne/Dentally/RiO). Establish NHS DSPT data governance. Configure RBAC roles (admin, clinician, approver). Run baseline evaluation on golden dataset. Confirm DCB 0129 clinical risk assessment.
2
Week 3–4: Core Agents Live
Deploy first 3 agents (highest-value use case for your setting). Wire HITL approval workflows in Temporal. Configure NeMo guardrails and patient data scrubbing. Set up Langfuse tracing and RAGAS eval gate. Clinician demo with real data.
3
Week 5–7: Full Agent Mesh
Deploy remaining agents. Configure Orchestrator routing. A/B test prompt variants. Enable drift detection. Train clinical staff on HITL workflow and approval interface. Clinical governance committee presentation.
4
Week 8–10: Production Hardening
Pen test + security scan. Load test to 3× peak concurrent users. Configure PagerDuty clinical alert routing. CQC/GMC compliance review. Caldicott Guardian sign-off. Produce runbook and clinical handover documentation. Go-live.