top of page
  • Writer's pictureAchint Sanghi


Perceiving several limitations of our WhatsApp chatbot, an android application called ‘Matar’ is combining the power of peer-to-peer networks with ASR and LLMs. As a part of the Agricultural Information Exchange Platform (AIEP) Matar is being piloted with communities in Bihar to understand more about how generative AI can be used to support rural women users.


Through the course of running the Bolbhav WhatsApp chatbot, a few limitations in the product came to light:

  • Literacy: though bolbhav was designed to be as accessible as possible, a basic level of literacy was required to use the chatbot independently

  • Smartphone ownership: Though WhatsApp is supported on smart feature phones, not all UX components (such as buttons and dropdowns) are supported. This makes the chatbot usage cumbersome or impossible for users without a smartphone.

  • Unidirectional flow: Bolbhav chatbot had a single flow for user feedback. Many users would ask questions as well as contribute information in this flow, but because of the UX of a chatbot, it was not possible to build any easy discovery feature of the content amongst users.

  • Gender biased: In the Indian context, decisions about the sale of crops are traditionally made by men. Thus, bolbhav’s userbase was entirely male and there was little scope to address the needs of women users.

We brainstormed the following product features to address each of these problems:


Product feature


Voice and audio based


App for JioPhone and KaiOS (Smart feature phones)

Unidirectional information flow

Peer-to-Peer Q&A flow

Gender bias

Thematic areas relevant to women and channels specifically for women’s communities

Experiments in AI

At the same time as our ideation exercise, APIs for LLMs such as GPT3.5 were becoming widely available for commercial adoption. While our early experiments with these LLMs indicated their severe lack of hyperlocal knowledge and information, they did provide instant replies with helpful generic information.

How do LLMs fare in addressing information needs of rural communities?




Hyperlocal context or realtime information

LLM generally cannot provide a satisfactory or useful answer

“How much does the CSC in my block charge to enroll in PM Kisan Yojana Scheme?”

Generic and objective information

LLM provides helpful context

“What documents are required to enroll in PM Kisan Yojana Scheme?”

As a compliment to peer-generated responses, we felt it could be helpful to incorporate AI-generated responses into the Peer-to-Peer Q&A flow.

Why “matar”?

In coming up with a name for the product, we wanted to communicate the idea of voice-based communication in a community. One idea was “utter”, meaning “To give public expression to; to speak”. An evolution of this word that would be easy to pronounce in Indian languages is “matar”, which means “peas” in Hindi.

Product development and pilot


There are approximately 100 million feature phone users in India. The JioPhone is among the popular devices in this category and runs on a version of the KaiOS operating system. We hypothesized that more rural women would have access to a feature phone than a smartphone.

Thus, our initial development effort of matar was focused on developing a KaiOS app for release on the JioPhone app store. However we faced a number of hurdles in developing for the JioPhone ecosystem, which led us to decide to focus on developing an android application first.

Challenges with JioPhone app development

  • The Jio developer ecosystem is still in its nascent stage. Documentation and guidelines are minimal and there are only a handful of developers who have experience developing KaiOS applications.

  • Due to the limited hardware capability of these devices, the app ecosystem is highly regulated by Jio, and the average app submission time can be between 2 to 6 months to accommodate for extensive testing. Regular app updates and product iterations are impossible for 3rd party apps on Jio.

  • Given that the JioPhone can only host a handful of applications, the Jio Developer Team prioritizes supporting well-known applications and Jio-branded apps. Having traction on the Android application would be helpful before submitting the JioPhone store.

An early prototype design of matar for JioPhone

💡 For a detailed explanation on the features and architecture of the matar android application, check out our extensive product documentation here.

Pilot and initial learnings

To gauge early user feedback, Matar was shared with 750 Bolbhav users of which around 50 registered on the app. Engagement on the app was low, with anywhere between 5 to 15% of the Daily Active Users creating a post on their first day of joining. In order to build a product that was useful for women users, we also conceptualized pilots of Matar with nonprofit groups that work with women. Our understanding is that significant work must be done in community development in order for the product to be a success.


In October of 2023, GIZ in partnership with FAIR Forward launched an initiative to create an MVP of an Agricultural Information Exchange Platform. The project is structured as a grant for organizations to develop a product that will:

  1. Cater to the needs of low-literacy and low-digital skill groups

  2. Be extensible across domains, geographies, and contexts

  3. Provide high-quality. diverse, personalized, and dynamic information and the capacity to engage in two-way communication

Matar is currently part of a cohort that includes IRRI, IFFCO-KISAN, and Dexian, and is running the below experiments:

  • Using a RAG pipeline with the LLM to support topic areas relevant to women (such as kitchen gardening and cattle rearing)

  • Tuning ASR models to facilitate use in local dialects such as Bhojpuri

  • Working with JEEViKA Didis to understand how matar can help support their knowledge extension to women, and incorporating feedback from women to improve the UI.

About the author: Achint is CTO at Gramhal.


Recent Posts

See All


bottom of page