About Random Politik

Transparency and information about the website

AI-Generated Content

IMPORTANT: All content on this platform — including quiz questions, evaluations, party comparisons, and political reports — is generated using LLM models from Mistral.
  • Responses may contain errors, inaccuracies, or outdated information
  • AI-models may have biases or misinterpret political nuances
  • Results should NOT be considered political advice
  • Always verify information with official party platforms

This platform is an experimental project and does not replace personal political research.

About the Project

Random Politik is an independent, experimental project that aims to use AI to help citizens explore the positions of Danish political parties.

The platform offers four tools:

  • Partitest — A dynamic quiz that matches your views with Danish parties
  • Politik Udforsker — Explore political areas and party positions across all categories
  • Parti Udforsker — View a single party's overall political proposals and positions, filtered by category
  • Politikrapport — Get a tailored AI-generated report on parties' positions on a topic of your choice

The project is non-partisan and operates without political affiliation.

Methodology

Why focus on parties rather than candidates?
Almost all political parties vote as a bloc in parliament on every decision — meaning the choice of party has far greater practical impact than the choice of individual candidate. Finding the party that best represents your views across issues is therefore the most effective starting point for navigating Danish politics.

How are questions generated?
Questions are selected procedurally across party policies within a chosen category. The selection favours proposals from the party most aligned with your previous answers, while avoiding questions already answered — so the quiz continuously adapts and uncovers new nuances.

How are clusters and similar policies identified?
All policies are vectorised using Mistral's embedding model and compared via cosine similarity. UMAP is applied for dimensionality reduction and HDBSCAN for clustering, automatically grouping thematically related policies that can be searched across parties.

Limitations & Disclaimers

  • The scoring model is approximate: It uses LLM evaluation of your answer against each party's political proposals. Results are indicative, not definitive.
  • Political positions are complex: A party may have many internal currents that a quiz cannot capture.
  • Data may be incomplete: Not all political proposals are necessarily represented in the dataset.
  • Results are NOT political advice: Use them as a starting point for your own research, not as a voting guide.

Technology & Data

  • Frontend: Next.js, TypeScript, CSS Modules
  • Backend: FastAPI (Python), Azure Cloud
  • AI: Models from Mistral ('mistral-small-latest' and 'mistral-medium-latest').
  • Data: Official party websites and programs (scraped and processed)
All internal data processing is done in English after analyzing the Danish websites. User-facing output is translated to the selected language by AI.

Contact

Do you have questions, suggestions, or feedback about the platform?

Get in touch here:
kai.zhang@round.dk