AI Advocate for the Poor

InspirationAI Advocate for the Poor

Justice should not depend on wealth, cognitive abilities, health, disability, faith, sexual orientation, or any other personal status.

Millions of people face decisions from courts, authorities, insurers, employers and other institutions without being able to understand the evidence, procedural history or available next steps. AI Advocate for the Poor turns complex documents into a transparent, source-grounded map that ordinary people can inspect and challenge.

This is a demonstration prototype, not a lawyer and not a substitute for professional legal advice.

What it does

The prototype:

analyzes a prepared Czech legal case study; reconstructs documents, events, proceedings and contradictions; separates extracted facts, legal interpretation, uncertainty and recommended next steps; supports factual conclusions with exact source quotations; distinguishes procedural actions from confirmation of wrongdoing; preserves human review instead of presenting AI output as a verdict; processes the demonstrated document workflow locally in the browser.

The live prototype also shows how one document can affect multiple proceedings without losing its source, date or procedural meaning.

Why it matters beyond one legal case

The Czech case is a demanding crash test, not the limit of the project. The same evidence-first architecture can support people dealing with public authorities, family and social matters, insurance claims, workplace disputes, consumer problems, journalism and nonprofit counselling.

The project can create public value while developing sustainable services for individuals, lawyers, NGOs, media organizations and institutions. A free public-interest layer can be supported by paid professional workspaces, integrations, document processing and institutional deployments.

Built during OpenAI Build Week

During Build Week we transformed an early visual concept into a safer working prototype:

introduced a prepared-sample boundary; prevented unsupported documents from receiving the sample’s legal interpretation; separated facts, interpretation, uncertainty and next steps; added quotation-backed conclusions; added regression tests against dangerous procedural confusion; expanded the case and proceeding map; created Czech and English public interfaces; improved documentation, deployment and repository hygiene.

The current release passes 188 automated tests.

How we built it

Codex was used as an engineering and verification partner for repository review, implementation, testing, documentation and deployment. GPT-5.6 supports the structured reasoning approach behind the prototype. The system is designed around four principles: source fidelity, procedural context, explicit uncertainty and human responsibility.

A living proof

After the competition deadline, development will continue in a clearly separated live layer. The next planned innovation is a public document intake: anyone will be able to submit a document and receive a quotation-backed assessment of its possible relevance, without being promised a legal outcome.

A conditional 2027 field pilot will accompany the “Peace of Cannabis for Everybody” journey from Geneva to Santiago de Compostela, approximately 2,000 kilometres, testing whether an AI-supported evidence map can remain understandable and useful outside a laboratory setting.

Links

Live application: https://dusandvorak-byte.github.io/ai-advocate-for-the-poor/

English version: https://dusandvorak-byte.github.io/ai-advocate-for-the-poor/en/

Source code: https://github.com/dusandvorak-byte/ai-advocate-for-the-poor

AI Advocate for the Poor does not promise victory. It makes evidence, uncertainty and institutional responsibility visible.

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for AI Advocate for the Poor

Built With

• codex
• css
• gpt-5.6
• html
• javascript
• openai
• pdf
• python