HOW IT WORKS
Record Cough
Caregiver records child's cough via WhatsApp in Bangla.
Audio Preprocessing
AI removes noise, segments cough, checks audio quality.
AI Respiratory Analysis
Wav2Vec2-XLSR-53 analyzes patterns from thousands of samples.
Severity Detection
Six-class pediatric classification with confidence scores.
Medical Knowledge Retrieval
RAG retrieves WHO IMCI & DGHS protocols for predicted condition.
Guidance & Referral
Bangla audio guidance & alert sent to nearest health worker.
AI ARCHITECTURE OVERVIEW
Modular, cloud-native backend for scalable pediatric care.
User Interaction
(WhatsApp · API)
Audio Processing
(librosa · ONNX)
AI Intelligence
(Wav2Vec2)
Knowledge Retrieval
(RAG · pgvector)
Decision Layer
(rules-gated)
Agent Orchestration
(MCP · LangGraph)
Data Infrastructure
(Supabase · Modal)
Scalability
(Continuous training)
Empowering health workers with AI precision.
Our technology is designed for the frontlines of rural healthcare, providing immediate, actionable insights where they are needed most.
Clinical decisions are deterministic per lib/claude.ts::decideSeverityMultiModal. No runtime LLM in the caregiver path — see ARCHITECTURE.md §3 carve-out.