Promise Tracker

Democracy Track is an AI-assisted civic research tool that lets citizens investigate politicians' promises against public record — with full source transparency, so you can decide for yourself.

Comment

Democracy Track — Promise Tracker

Track 4: Governance & Collaboration Claude Builder Club Hackathon @ Imperial College London, Spring 2026

What did they promise? What did they deliver?

Democracy Track is an AI-assisted civic research tool that lets citizens investigate politicians' promises against public record — with full source transparency, so you can decide for yourself.

The Problem

Democratic institutions are strained. Citizens can't easily track whether politicians deliver on their promises. Information is fragmented across news outlets, parliamentary records, and fact-checkers. Polarisation makes it hard to separate rhetoric from reality.

Who faces this problem:

  • First-time voters who want to understand what politicians actually delivered before casting their vote
  • Citizen journalists who need a starting point to investigate local representatives
  • Civic organisations tracking accountability across multiple politicians

What We Built

A full-stack SvelteKit application deployed on Cloudflare that combines live news data with AI analysis to create structured, source-transparent politician profiles.

Architecture: Real Data Pipeline (not a chatbot wrapper)

User enters politician name
        |
        v
Step 1: Google News RSS --- fetch 15 live headlines
        |
        v
Step 2: Claude API --- analyse promises against real news + public record
        |
        v
Step 3: Structured output --- score breakdown, timeline, indices, connections

This is critical: the AI does not generate news or invent facts. It receives real headlines fetched from Google News and analyses them against known public record. Every news item links to its real source URL.

Key Features

Promise Fulfillment Index (PFI)

A transparent, auditable scoring methodology adapted from PolitiFact and FullFact:

  • Weights: Kept (1.0), Partially Kept (0.5), In Progress (0.3), Broken (0.0)
  • Unverifiable promises excluded from the scoring denominator
  • 4 adjustment factors (source diversity, recency, promise significance, contextual factors), each +-5 points
  • Full arithmetic visible on hover — judges can audit every number

Political Standing & Opposition Mapping

  • Left-right spectrum visualisation with politician and opposition markers
  • Policy Overlap Score: reveals when opponents have similar actual policies despite public rhetoric (exposes "theatrical opposition")
  • Coalition allies and key disagreements mapped

Country-Level Context

  • Corruption Perception Index (Transparency International)
  • Democracy Index (Economist Intelligence Unit) with regime classification
  • Press Freedom Index (Reporters Without Borders)

These provide essential context: a politician's record must be interpreted within their country's institutional environment.

Memorandum

A diplomatic-style briefing: rise to power, political philosophy, achievements, controversies, trajectory. Gives full context to someone with no prior knowledge.

Real News with Proximity Scoring

  • Live headlines from Google News RSS
  • Each news item scored by proximity to a specific promise (0.0-1.0)
  • Sorted by relevance, linked to source URLs
  • Connected to specific promises via related_promise_index

Timeline of Key Events

  • 8-15 key political events: meetings, legislation, speeches, elections, scandals, appointments
  • Each event lists people involved and outcomes
  • Categorised and colour-coded by type

Connection Subsearch

After the initial analysis, users can investigate connections to any person, organisation, or business:

  • Fetches joint news for both entities
  • Maps key links by nature (political, financial, lobbying, personal)
  • Identifies shared affiliations, financial ties, risk notes
  • Shows timeline overlap — events where both were present
  • Connection strength ring (0-100%)

Politician Status Awareness

Adapts dynamically: serving, retired, resigned, removed, lost election, deceased, suspended, term ended — each with contextual detail and appropriate labelling.

Multi-Language Support

20 languages including English, Russian, Kazakh, Ukrainian, French, German, Spanish, Arabic, Chinese, Japanese, Korean, Turkish, and more. All analysis content is generated in the selected language while maintaining structured data integrity.

Print-Ready

Full print stylesheet: dark theme inverts to white background, interactive elements hidden, all data preserved with readable formatting.

Tech Stack

Layer Technology
Frontend SvelteKit 2 + Svelte 5 (runes)
Backend SvelteKit server actions
AI Claude API (claude-sonnet-4) via REST
Live Data Google News RSS
Hosting Cloudflare Workers + D1
Language TypeScript

Ethical Considerations

What could go wrong

Weaponisation risk: This tool could generate misleading screenshots about politicians. Every result carries a prominent AI-generated disclosure banner and links to verify claims independently.

Oversimplification: A left-right spectrum cannot capture political nuance. The "policy overlap" metric helps reveal when opponents are closer than rhetoric suggests, but real politics is multi-dimensional.

Whose voices?: English-language sources dominate. AI training data has Western bias. The tool works best for well-covered politicians and may miss perspectives from underrepresented communities. Country indices (press freedom, democracy index) provide context for this limitation.

Empowerment, not replacement: This tool helps people start researching their representatives — it does not tell them what to think. The goal is informed citizens, not automated verdicts.

Safeguards built in

  • Prominent AI disclosure banner on every result
  • Methodology note with honest limitations
  • Confidence level indicator (high/medium/low) with reasoning
  • Unverifiable claims excluded from scoring
  • Counter-evidence required for each promise assessment
  • Factual accuracy rules for non-Western politicians (lower confidence when data is thin)

Critical Questions (from hackathon brief)

Who are you building this for and why do they need it? First-time voters, citizen journalists, and civic organisations who need structured, source-transparent information about politicians' track records — not opinions, not spin, but evidence they can verify independently.

What could go wrong and what would you do about it? The tool could be used to generate negative profiles for manipulation. We address this with: prominent AI disclaimers, source URLs for independent verification, balanced scoring methodology that presents both achievements and failures, and honest confidence ratings when data is limited.

How does this help people rather than make decisions for them? The tool presents structured evidence and says "decide for yourself." The scoring methodology is fully transparent and auditable. The memorandum provides context without editorial framing. Every claim cites sources. The user is always the judge.

Team

Kassymkhan Sundetbay Bakhtiyar Mautov

Claude Builder Club @ Imperial College London Track 4: Governance & Collaboration Spring 2026 Hackathon

Built With

  • claude
Share this project:

Updates