Baby Pulmo
Try Demo

HOW IT WORKS

1

Record Cough

Caregiver records child's cough via WhatsApp in Bangla.

2

Audio Preprocessing

AI removes noise, segments cough, checks audio quality.

3

AI Respiratory Analysis

Wav2Vec2-XLSR-53 analyzes patterns from thousands of samples.

4

Severity Detection

Six-class pediatric classification with confidence scores.

5

Medical Knowledge Retrieval

RAG retrieves WHO IMCI & DGHS protocols for predicted condition.

6

Guidance & Referral

Bangla audio guidance & alert sent to nearest health worker.

AI ARCHITECTURE OVERVIEW

Modular, cloud-native backend for scalable pediatric care.

01

User Interaction

(WhatsApp · API)

02

Audio Processing

(librosa · ONNX)

03

AI Intelligence

(Wav2Vec2)

04

Knowledge Retrieval

(RAG · pgvector)

05

Decision Layer

(rules-gated)

06

Agent Orchestration

(MCP · LangGraph)

07

Data Infrastructure

(Supabase · Modal)

08

Scalability

(Continuous training)

WHO guided · IMCI alignedBangladesh focused · DGHS protocolsBangla Voice Support

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.