How AI Works (Doctor‑First)
Faster decisions for doctors, safer collection for patients, lower ops overhead for your team.
Televisit summary in minutes
Auto‑escalation for red flags
Smart recheck windows
Routing cuts late visits
Multilingual WhatsApp education
Televisit‑Ready Summary NLP/RAG
- Turns home data + labs into a 6‑part clinical brief
- Flags red‑flags; cues guideline‑aligned actions
- Models: Claude/GPT via Bedrock; RAG over OpenSearch
Recheck Recommender Policy + ML
- Suggests next TSH/INR/HbA1c window
- Explains rationale; defaults conservatively
- Rules + LightGBM; MD confirmation required
Triage & Escalation Rules + LLM
- INR>5, eGFR↓, SpO₂<90%, SBP>180 → ping MD
- Prefilled message; MD confirms next step
- EventBridge → SNS/WhatsApp
Routing Optimizer OR/ML
- Assigns nurse by skill, proximity, priority
- ETA model reduces late arrivals
- OR‑Tools + XGBoost in SageMaker
Patient WhatsApp GPT Education
- Doctor‑approved FAQs only; bilingual
- Deflects routine questions
- Refuses diagnosis; escalates to MD
Clinician Copilot RAG
- Search prior orders; draft notes
- Pulls Indian guidelines snippets
- RBAC + PHI tokenization
AWS Implementation Flow
Reference stack with Bedrock, SageMaker, EventBridge, DynamoDB, S3, Redshift.
Safety, Privacy & Compliance
- PHI tokenized before LLM calls; KMS‑encrypted keys
- RBAC on vector store; TTL for embeddings with PHI
- Doctor confirmation required for any therapy suggestion
- Audit trails via CloudWatch & Lake Formation