Interactive Democracy

Civic infrastructure for the ASI era. A personal civic agent - on your device, answering only to you - helping every person weigh public decisions in minutes, from street to nation, into the future.

Comment

Inspiration

Most people are asked for opinions before they have had the opportunity to understand an issue. Every poll, consultation and comment section works that way, and it produces heat instead of light. Democracy has an education problem before it has a voting problem.

Meanwhile, the most sophisticated preference-learning machinery ever built, cookies, histories, inferred profiles, has spent two decades learning what each of us values, and pointed all of it at selling. Nobody pointed it at citizenship.

A new generation is growing up inside the AI era. For them, AI will not be a new tool; it will simply be how the world works. As AI reshapes work, public services, information and public decision-making, civic infrastructure must evolve just as fast and it must learn people the way the best technology does, while answering only to them.

Interactive Democracy is our answer: civic infrastructure for the AI era, and the ASI era ahead.

What it does

Interactive Democracy reverses the usual sequence: understand first, then decide.

Example: a resident has two minutes to decide whether their borough should fast-track affordable housing. The agent summarises the issue, who is affected, the strongest argument for building, the strongest argument against, and the evidence confidence level. The resident gives a leaning, sees the strongest good-faith alternative, and submits an informed contribution, all in under two minutes.

For any public decision, from a damaged pavement on one street to national food policy, the platform assembles a neutral Civic Intelligence Brief: stakeholders, evidence with explicit confidence levels and sources, competing viewpoints each shown with its own weakness, trade-offs, and honest uncertainty. Citizens choose how much time they have (a guided two-minute path or the full brief), form a leaning, and then meet the heart of the platform: a constructive deliberation step. Instead of being argued with, the citizen is guided, their position is acknowledged and shown to be credible, the strongest good-faith alternative is presented with its best evidence, the honest trade-off between the options is named, and only then are they asked whether anything shifts. Any answer is a legitimate outcome. The tone is civic guidance, not debate: the system never advocates a position, never argues, and never tries to "win", it exists to make the citizen's own choice better informed.

Each citizen also has a personal civic agent. It learns their values entirely on-device (never uploaded, fully visible, erasable in one tap), pre-reads every open decision and drafts where they would likely stand, but nothing is ever submitted without explicit human confirmation, and never before the strongest alternative has been shown. Consent-gated autopilot: the agent does the work; a human always fires the shot.

Decision-makers never see individuals, only structure: how views distribute, the values people expressed, their concerns, novel positions the framing missed, and the nuance metric, the share of participants who refined or changed their position after weighing the strongest alternative. Not consensus, better understanding before opinion.

Six decisions run live today across six civic scales, neighbourhood, ward, constituency, borough, region, nation, on the same engine. Opening a new decision requires one JSON configuration file.

How we built it

Interactive Democracy is not a civic platform with an AI chatbot added on top. Gemini is not a chatbot layer inside this product; it is the operating engine. It performs the research, structuring, deliberation support and contribution analysis that make informed participation scalable, humans remain responsible for judgment, consent and final submission. Every brief is researched, structured and sourced by Gemini (JSON mode, with response schemas derived from our zod schemas, a single source of truth, validated on receipt). Every constructive-deliberation step is composed live, specific to the citizen's stated leaning. Every contribution is summarised, value-tagged, privacy-scrubbed and classified by Gemini before storage.

Stack: Gemini API (gemini-2.5-flash, JSON mode) · Cloud Run (europe-west2) · Firestore · Next.js / TypeScript. Contributions are stored under anonymous rotating session IDs; there is no code path from the dashboard to an individual contribution, privacy is enforced by architecture, not by promise.

Challenges we ran into

The hardest problem was tone. Our first deliberation step presented "the strongest case against your view", accurate, but it felt like debate, and debate makes people defensive. We restructured it into a constructive sequence: acknowledge the citizen's position, show why it is credible, present the strongest alternative, name the trade-off, then ask an open question. Same intellectual honesty, entirely different experience.

What we learned

People engage with opposing evidence when their own position is taken seriously first. Measured as the nuance metric, that willingness to refine is the real output of the platform and it is measurable, which most civic education is not.

What's next:

The personalisation advertising never attempted

Advertising built the most sophisticated preference-learning machinery in history and pointed every bit of it at selling. Interactive Democracy points that depth of understanding the other way. Your civic agent learns what you value because you show it, openly: on your device, never uploaded, fully visible, erasable in one tap. It answers to no advertiser, no party, no platform - only to you.

Today it learns only from explicit in-app choices. It deliberately uses no cookies, no browsing history, no inferred advertising-style profiles. For a civic system handling political opinion, that is the correct default - and it will remain the default. With explicit, revocable consent, future versions could add:

  • consent-based user profiles, on-device by default

  • privacy-preserving civic personalisation (on-device inference; aggregation and differential-privacy techniques wherever any signal leaves the device)

  • local issue relevance- surfacing the decisions that actually affect you: your street, ward or borough

  • multilingual and age-appropriate explanations of every brief

  • adaptive civic learning that meets each person at their current knowledge level

  • issue recommendations based on the citizen's interests and their community's needs

  • transparent control over every signal used - visible, adjustable, erasable

  • a counter-manipulation shield - with consent, the agent warns you when content you encounter about a live decision carries the fingerprints of targeted persuasion: personalisation as your defence, not your exposure

  • and by design: no hidden manipulation, no political targeting, no advertising-style exploitation

The core principle is fixed: personalisation exists to increase understanding, agency and participation, never to influence opinion, manipulate behaviour or extract attention.

Business model

Interactive Democracy is sustainable civic infrastructure, not an advertising business. Initial customers are local authorities, civic institutions, schools, NGOs and public-interest organisations that need better participation, consultation and civic learning tools. The first revenue path is paid pilots: opening a specific public decision to informed input, producing aggregate insight dashboards, and supporting institutions with evidence-based participation reports. Longer term, this expands into subscriptions, education licences, white-label deployments for cities and regions, institutional dashboards and international civic pilots. Revenue scales with public value delivered, participation, understanding, institutional adoption, never with attention captured or data extracted. Individual-level data is never sold and never used for targeting.

Every pilot closes the loop. The commissioning institution publishes a reasoned public response to the top concerns and novel positions its citizens raised, so participation visibly lands, and institutional responsiveness becomes part of what the platform measures. Civic technology fails when contributions disappear into a void; here, the response is part of the product.

The ambition

Elections ask for a decision every few years; the decisions that shape lives are made every day. Our ambition is to give billions of people, starting with the generation that will live longest with AI, a meaningful way to understand, deliberate and contribute continuously, and to give institutions collective intelligence instead of noise.

Civic infrastructure at that scale has to be built differently, and we hold ourselves to that standard now. Privacy is enforced by architecture, not by promise, and neutrality will be too. Our roadmap commits to briefs composed by multiple independent models with their disagreements surfaced as contested framing, a public provenance log for every brief, and publishing the brief format and deliberation sequence as an open standard that any city or country can implement and audit. A protocol, not a platform: infrastructure this important cannot be a black box owned by anyone - including us.

It must also serve the people who are not in the room. Every brief will carry a future-generations view, the honest 10-, 50- and 100-year trade-offs of the decision, because those who will live longest with each choice deserve a seat in it. Deliberation will extend from citizen-and-agent to citizen-and-citizen, with the agent facilitating small groups toward common ground rather than victory. And the aggregate record of what people understood, valued and decided becomes a permanent public archive for the generations that follow.

This is what civic infrastructure for the AI age looks like: participation that scales, consent built into the architecture, a citizen who is more informed after every single use, and a platform whose own roadmap will, as soon as enough citizens are here, be decided on the platform itself.

Built With

Share this project:

Updates

Submission history