A community health worker in a remote village sits with an elderly patient. She pulls out her basic phone, dials a number, and asks in Bangla: “রোগীর সুগার ৮.৫, এখন কী করব?”. A calm voice responds, walking her through the Ministry of Health's protocol — adjust diet, increase physical activity, and schedule a follow-up in two weeks.

This is the Bangladesh Diabetes Assistant, a groundbreaking voice AI tool built by Medtronic Labs and Gooey.AI, and it's already changing how diabetes care is delivered in underserved communities.

The diabetes challenge in Bangladesh

Bangladesh has one of the fastest-growing diabetes burdens in the world. An estimated 1.3 crore (13 million) adults live with diabetes, but many are undiagnosed or poorly managed. The ratio of diabetes patients to endocrinologists is staggering — often 10,000:1 in rural areas.

Community health workers (CHWs) are the frontline, but they often lack access to up-to-date clinical guidelines. Printed manuals are outdated, and training is infrequent.

Enter the Diabetes Assistant

The Bangladesh Diabetes Assistant is a voice‑based AI tool designed specifically for CHWs and patients. It's built on three core principles:

• Grounded in权威 guidelines: All answers come from the Ministry of Health's diabetes management protocols and international standards adapted for Bangladesh.
• Voice-first: Works on any phone — no smartphone, no internet required.
• Bangla-native: Understands regional dialects and code-mixed speech.

The technology: RAG + Ministry guidelines

Like KrishokBondhu in agriculture, the Diabetes Assistant uses Retrieval-Augmented Generation (RAG) to ensure accuracy. The knowledge base includes:

• Ministry of Health's National Guideline for Diabetes Management (2024 edition).
• WHO diabetes protocols adapted for Bangladesh.
• Drug formularies with local brand names.
• FAQ database from diabetes clinics.

When a CHW asks a question, the system retrieves the most relevant sections and generates a response grounded in those sources — no hallucinations, no made-up advice.

Beyond CHWs: patient self-care

The assistant also has a patient-facing mode. A diabetes patient can call and ask:

“আজ সকালে খালি পেটে সুগার ৭.২, এটা কি স্বাভাবিক?”

The assistant responds with personalized advice based on the patient's history (if available) and general guidelines. It can also remind patients about medication times and clinic appointments via outbound calls.

How Speaklar's infrastructure powers it

The Diabetes Assistant runs on Speaklar's voice AI platform:

• Bangla ASR: Converts CHW queries to text, trained on healthcare conversations.
• RAG pipeline: Retrieves from the Ministry of Health knowledge base.
• Bangla TTS: Speaks responses in a clear, warm voice optimized for medical terminology.
• Telephony integration: Works on any phone via a short code (e.g., 1627).

The entire system is HIPAA-inspired compliant (patient data encrypted, no recordings stored without consent).

Real voices: what users say

Scaling to other NCDs

The success of the Diabetes Assistant has sparked interest in similar tools for:

• Hypertension: Blood pressure management protocols.
• Maternal health: Pregnancy care guidelines.
• Child vaccination: EPI schedule reminders.

Each would use the same RAG + voice template, with knowledge bases drawn from relevant Ministry guidelines.

Challenges and lessons

• Medical terminology in Bangla: Translating terms like "insulin resistance" into plain Bangla that patients understand required careful design.
• Fallback to human: For complex cases (e.g., "রোগী অজ্ঞান হয়ে গেছে"), the assistant immediately transfers to an emergency hotline.
• Trust building: Initial users were skeptical — "এটা কি ডাক্তারের মত সঠিক বলবে?" — but accuracy over time built confidence.

The future: integrated with national health systems

Plans are underway to integrate the Diabetes Assistant with the government's Health Information System (HIS). When a CHW counsels a patient, the interaction could automatically update the patient's electronic health record. This would give doctors a richer picture during consultations.

🩺 সঠিক তথ্যই সঠিক চিকিৎসা — কণ্ঠে এলো স্বাস্থ্যসেবা

---