A farmer in Rangpur notices yellowing leaves on his paddy field. He pulls out his basic mobile phone, dials a number, and asks in his native dialect: “ধান গাছের পাতা হলুদ হচ্ছে, কী করব?”. A calm voice responds in Bangla, explaining the likely cause (nitrogen deficiency) and suggesting the right fertilizer dose — all pulled from official agriculture department manuals.

This isn't a pilot. It's KrishokBondhu, Bangladesh's first RAG‑powered AI voice assistant for farmers, and it's changing how agricultural extension services reach rural communities.

The challenge: 16 million farmers, too few extension officers

Bangladesh has over 16 million farming households, but only a few thousand government extension officers. Most farmers never get expert advice when they need it. They rely on local pesticide dealers, whose advice often leads to overuse of chemicals and lower yields.

Smartphone penetration is rising, but rural internet sits at only 46%. Voice is the only channel that reaches every farmer, regardless of literacy or connectivity.

What is RAG and why does it matter?

RAG = Retrieval-Augmented Generation. It sounds technical, but the idea is simple: instead of letting an AI guess answers from its general training (which can cause "hallucinations"), RAG first retrieves relevant facts from a trusted knowledge base, then generates a response grounded in those facts.

KrishokBondhu's team took 2,500+ pages of scanned agricultural manuals, leaflets, and guides from the Department of Agricultural Extension. They used OCR to convert them to text, normalized the Bangla, and stored them in a vector database. When a farmer asks a question, the system:

• Converts speech to text (ASR).
• Finds the most relevant manual sections (retrieval).
• Generates a concise spoken answer in Bangla (generation + TTS).

72.7% high-quality response rate

In pilot evaluations, KrishokBondhu achieved a 72.7% high‑quality response rate — meaning farmers got accurate, actionable advice. For the remaining cases, the system intelligently escalated to a human expert or offered general information. This is far better than generic chatbots, which often fail on domain‑specific questions.

The key? Grounded retrieval. By anchoring answers in official documents, the AI never recommends a banned pesticide or an incorrect dosage.

How Speaklar's infrastructure enables this

KrishokBondhu is built on a stack that includes:

• Speaklar’s Bangla ASR & TTS — optimized for rural dialects and background noise (farmers often call from the field).
• RAG pipeline using a vector database of agricultural documents.
• Simple telephony integration — farmers call a local number; no app required.

The entire system is designed for low bandwidth — voice works on 2G networks.

Beyond queries: proactive advisories

KrishokBondhu also makes outbound calls for:

• Weather alerts: “আগামী ২৪ ঘণ্টায় বৃষ্টির সম্ভাবনা। ধান কেটে গোলায় তুলে ফেলুন।”
• Pest outbreaks: When the department detects an infestation, the bot calls farmers in the affected area with preventive measures.
• Government subsidy info: “সারের ভর্তুকি চলছে, নিকটস্থ ডিলার থেকে সংগ্রহ করুন।”

One pilot in Rangpur division reached 12,000 farmers in a single week — something human extension officers couldn't do in a year.

What farmers say

Scalability: from agriculture to every sector

KrishokBondhu proves that RAG + voice AI can work for any domain with a trusted knowledge base. Imagine:

• Healthcare: A voice bot that answers based on Ministry of Health treatment protocols.
• Legal aid: Grounded in Bangladesh Labor Law or land rights documents.
• Education: NCTB curriculum answers for students without internet.

This is the future of domain‑specific voice AI in Bangladesh.

How your organization can build a RAG voice bot

You don't need a team of AI researchers. Speaklar provides:

• Document ingestion pipeline (OCR, chunking, vector embedding).
• Bangla voice interface (ASR + TTS) fine‑tuned for local accents.
• Telephony integration — farmers call a regular number.

Whether you're in agriculture, healthcare, or finance, a RAG‑powered voice bot can deliver expert answers at scale.

🌱 কৃষকের কণ্ঠে এআই — ফসলের ফলনে নতুন দিগন্ত

---