The Dental Dispatcher

An AI voice assistant for dental clinics that analyzes X-rays, explains results, manages refill requests, tracks patient history, and delivers context-aware guidance through phone conversations.

Comment

Inspiration

Dental clinics deal with repetitive but critical conversations every day:

• “What does my X-ray show?” • “Is this urgent?” • “Can I get a refill?” • “What did my dentist say last time?”

Front desks are overloaded, dentists are interrupted, and patients are often anxious and confused. We wanted to build an AI system that could understand patient context, review clinical artifacts, and respond clearly and responsibly- over the phone.

What it does

The Dental Dispatcher is an AI-powered phone assistant for dental clinics.

When a patient calls the Amazon Connect number: 1. Lex manages the conversation. 2. A Lambda orchestrator (dentalnote-lex-orchestrator) retrieves patient context from DynamoDB. 3. Claude on Amazon Bedrock generates a structured, patient-friendly response.

It can:

• Explain X-ray findings in plain language and use structured label reports and overlay images for analysis • Reference patient history for context-aware responses • Handle automatic prescription refill requests • Provide urgency guidance and next-step recommendations

How we built it

1) X-ray Model Training and Analysis Flow

We trained a dental X-ray detection model to classify:

• healthy • caries • impacted • infection • fracture • broken crown or root

Pipeline: 1. Labeled X-ray dataset 2. Object detection training 3. Model outputs structured findings with confidence scores 4. Overlays generated for visualization 5. Reports stored in S3 6. Summaries stored in DynamoDB

Each X-ray produces: • Original image • Overlay visualization • Structured JSON label report • Aggregated findings (count + confidence threshold)

This structured output feeds into Claude for controlled explanation.

2) AI Voice Orchestration Flow

Phone Call → Amazon Connect → Lex bot → Lambda Orchestrator → DynamoDB (patient history + X-ray summaries) → Bedrock Claude → Response back to caller

We separate responsibilities:

• X-ray summarization Lambda • Voice orchestration Lambda • DynamoDB for structured context • Bedrock for natural language explanation

This modular design keeps clinical analysis separate from conversational orchestration.

Challenges we ran into

• Preventing hallucinations in medical-style explanations • Ensuring summaries strictly reflect structured findings • Handling low-confidence detections safely • Keeping phone responses short and calm instead of long paragraphs

Accomplishments that we're proud of

• End-to-end AI phone assistant working with real X-ray artifact data

• Context-aware conversations using patient history

• Structured summarization tied directly to model outputs

• Visual overlays linked to explanations

• Safe and consistent medical-style disclaimers

• Refill request handling logic integrated into conversation

What we learned

• In healthcare use cases, guardrails matter more than creativity

• Structured data + AI explanation is far more reliable than free-form prompts

• Voice UX requires clarity, brevity, and reassurance

• Confidence thresholds are critical in medical AI

• Architecture separation improves safety and scalability

What's next for The Dental Dispatcher

• Integrate appointment scheduling

• Enable live transfer to clinic staff

• Add SMS/email summary after call

• Add multilingual voice support

• Improve model training with more dental categories

• Add audit logging and compliance hardening

• Deploy pilot with real clinics

Future Scope of Visualization

Currently, we generate X-ray overlays with bounding boxes and structured findings. In future versions, we plan to expand visualization into a more interactive and clinically useful system.

Planned enhancements include:

• Color-coded bounding boxes based on confidence and severity

• Confidence heatmaps layered on X-rays

• Tooth-number mapping for structured dental chart integration

• Interactive dashboard with click-to-explain findings

• Historical comparison view across patient visits

• Risk scoring and urgency indicators

• Patient-friendly visual explanations in a portal

• Model performance and drift monitoring dashboard

These improvements will strengthen explainability, clinical trust, and real-world deployment readiness for The Dental Dispatcher.

Built With

Share this project:

Updates