Abstract
NexV is a dental practice management platform built on AWS with an AI-first architecture: every AI model runs through Bedrock or SageMaker, no external AI APIs, all processing in the practice's AWS account. It covers clinical charting, X-ray analysis, voice dictation, intra-oral photo AI segmentation, scheduling, patient engagement, and a 24/7 AI phone agent.
Problem
Dental practice software is expensive, slow to iterate, and architecturally hostile to AI integration. Adding AI capabilities to existing systems typically means bolting on third-party APIs that create data residency risks, latency problems, and dependency on external model providers.
A platform built for AI from the start can embed model inference directly into clinical workflows — X-ray analysis appears inline as the image loads, dictation parses directly to chart fields, and the phone agent uses the same Bedrock models as the rest of the application.
Approach
Platform architecture
The application runs on AppSync GraphQL with 91 resolvers (45 queries, 46 mutations), Cognito for staff and patient auth pools, and a single DynamoDB table with per-practice tenant scoping. Three SageMaker Serverless endpoints serve the dental detection pipeline: tooth detection, smile inpainting, and tooth segmentation.
Model assignments
Bedrock Opus 4.6 handles X-ray analysis, SOAP note transcription (Scribe), and smile simulation. Sonnet 4.6 powers the AI phone agent (via Amazon Connect + Lex V2), clinical assistant, call analytics, and scheduling AI.
Clinical workflows
FDI tooth notation (Australian standard, 11–48 permanent, 51–85 primary) displays throughout the UI; internal storage stays in sequential 0–31 arrays to avoid data migration.
Voice dictation uses Amazon Transcribe Medical for streaming transcription and Bedrock Sonnet for clinical parsing. The parse-dictation API converts free-form voice input to structured chart commands (tooth number, surface, condition) with a confidence score, and shows a preview for clinician confirmation before applying.
Status
- V-Pulse ($199/mo) and V-Sonic ($399/mo) tiers are priced and architected.
- 60+ pages, 57+ API routes, 91 GraphQL resolvers implemented.
- 363 official ADA item codes (13th Edition) and full FDI notation live.
- Dental detection model: v5-R2 YOLOv8m at 0.695 mAP50, 32 classes, deployed to SageMaker.
- FDI notation, voice dictation, and intra-oral photo AI segmentation shipped in PR #38.
- AI Phone Agent add-on (+$199/location/month) uses Amazon Connect + Lex V2 with Bedrock Sonnet.