Serverless WhatsApp at scale

A developer published a case study on April 19 showing how a WhatsApp farm advisor for more than 100 million Indian farmers was built entirely on serverless Amazon Web Services tools. (dev.to) The system, called AgriNexus AI, sends advice on WhatsApp and keeps following up until a farmer replies “हो गया,” or “done,” according to the post by Prasad Tilloo. The article says the service targets India’s last-mile farm advisory gap, where timing can matter more than the existence of advice itself. (dev.to) The architecture in the post uses Amazon API Gateway, Web Application Firewall, Lambda, Simple Queue Service First-In-First-Out queues, Bedrock, DynamoDB Streams, EventBridge Scheduler, Transcribe, and Polly. Tilloo wrote that the design avoids long-running state machines by scheduling one-shot follow-ups at 24 and 48 hours, then canceling them when a farmer confirms completion. (dev.to) Serverless systems run code only when an event arrives, instead of keeping servers running all day. Amazon says Lambda is designed for event-driven workloads in which services like API Gateway, Simple Queue Service, DynamoDB Streams, and EventBridge trigger short-lived functions as requests come in. (docs.aws.amazon.com) That model fits chat systems because message traffic comes in bursts, not on a fixed schedule. Amazon’s own WhatsApp guidance from October 31, 2025 recommends modular, event-driven designs using Simple Notification Service, Simple Queue Service, and Lambda to absorb spikes and keep downstream systems from being overwhelmed. (aws.amazon.com) The scale claim sits against a very large farm base in India. The Agriculture Census portal run by India’s Department of Agriculture and Farmers Welfare lists 146,453,741 operational holdings in the 2015-16 census and says the 2021-22 census is the 11th in the series. (agcensus.da.gov.in) Tilloo wrote that one cost decision was replacing OpenSearch Serverless with S3 Vectors because always-on capacity charges dominated early bills at low query volume. He modeled the resulting cost at about $0.54 per farmer per year at a 10,000-farmer scale, according to the post. (dev.to) The knowledge base in the case study pulls from Indian Council of Agricultural Research, Food and Agriculture Organization, and National Food Security Mission documents, and the post says answers show source links to farmers. Tilloo wrote that the retrieval-and-generate setup uses Amazon Bedrock with Claude and a vector knowledge base. (dev.to) The post was published the same day community voting opened for AWS Builder AIdeas 2025 finalists, with voting running April 18 through April 23 Pacific Time. In the article, Tilloo framed the project as both a production architecture example and an argument for reducing operations work in large public-facing services. (dev.to)