RachchaBanda

RachchaBanda is an AI-assisted system to accommodate more nuanced public discourse and opinion sharing on matters related to the local community. Its aim is to make grassroots institutions more democratic and inclusive, to expand local decision-making in time and space, to give individuals and marginalised groups a secure space to express an opinion, and to help local officials take better-informed decisions.

Why

Whether it is the council of elders in Africa or the gram panchayat in India, local communities have long built ingenious mechanisms for solving their own problems through discussion. Those institutions carried the same faults as the societies around them — class hegemony, social inequality, rigid hierarchy — in whose voices they let into a decision.

Centralisation since the rise of the modern nation state reduced many of them to symbols. The existing panchayat system does have a mechanism for capturing opinion, through ward members, a secretary and a sarpanch. But representatives are elected once every four years along state party lines, without much consideration of local issues at the time of voting, and once elected there is little that brings the community back into a panchayat decision.

This project addresses the operational limits that keep communities out of that process.

How it works

At the core is an audio recording device, also called RachchaBanda, which is circulated among community members before a vote on a particular issue. It captures discussion — by individuals, small groups or large groups — at people’s own convenience and in privacy. Conversations in the local language are then processed by an AI system set up in the panchayat office itself.

1. The panchayat secretary and sarpanch formulate a problem, with preliminary information, related discussion topics and a decision deadline.
2. Devices are circulated through ward members to conduct discussions, at whatever granularity happens organically.
3. Conversations are processed near-real-time at the panchayat office while the devices are still in circulation.
4. A day before voting, the community is shown the high-level patterns: the dominant approach supported, the dominant approach opposed, where agreement held and where it did not, and how those patterns break down by ward, gender, class and caste — with the anonymity of individuals and groups preserved.
5. On the day, people vote knowing both their own discussion and everyone else’s.

Why AI is central

The mechanism could in principle run without AI, but only with heavy manual involvement — and that involvement would itself create a new class of people with the power to shape the outcome. AI does the collation, infers the patterns and themes, and makes them consumable by the community, without a human gatekeeper in the middle.

A worked example: waste water

A village with no waste water treatment infrastructure, where households use septic tanks or divert to open canals, needs a plan. The problem is broken into discussion topics that build on each other: do we need more information first; centralised or decentralised technology; local people or an external contractor; how to choose between contractors; how to monitor and verify the work; how to distribute work among locals if we do it ourselves.

Each discussion is processed for what it revolved around, what everyone agreed on, what they did not, what solutions came up, whether a dominant voice biased the direction, and what share of the group actually spoke.

Openness and sustainability

All project documentation, including the codebase, is open-sourced, and the panchayat joins a data collective that makes community data available under appropriate terms.

Sustaining it rests on three things: autonomy over which issues get discussed stays with the community; people who give up working time to talk need real incentives to do so; and funds have to be devolved to panchayats as policy, on the strength of these being more transparent and accountable institutions. Day-to-day running is funded from local government revenue, including local taxes and data-collective revenue, with a periodic audit-and-improve cycle based on what the field reports.

Open problems

The hard parts are not mostly technical. Adoption. Giving enough information to inform a discussion without steering it. Keeping discussion topics current enough to move a decision forward. Identifying and weeding out vested interests and coercive pressure. Designing incentives that make local power-holders allow open discussion at all. Keeping opinions genuinely anonymous.

The technical ones are real too: an LLM suited to this specific purpose, training and inference that a panchayat can afford, robust privacy, officials able to prompt the system usefully, and a local talent pool that can maintain the infrastructure.