BEFORE THE TEST
Drafts dated 18 July 2026
The new referral notice had not yet been safely propagated through all three documents, and some wording treated a prepared draft as already filed or evidenced.
PROJECT MISSION
From thousands of fragmented items to an auditable case
A person with limited money or health may have a legitimate claim and enough evidence, yet still fail because they lack the resources, time, or professional capacity to organize their case.
AI Advocate is designed to compensate for this disadvantage. It turns fragmented legal and life materials into a structure of people, institutions, proceedings, events, claims, and evidence. It can then gather the relevant sources within a selected branch and prepare a proposed next step for human review.
fragmented materials→auditable structure→relevant branch→proposed action
This does not replace a lawyer. It helps a person reach the point where a lawyer, legal-aid organization, ombudsperson, or public authority can actually help.
AUDITABLE MEMORY
Statements, institutions, links, and case law
The public case-study author is named with consent. Contact details and irrelevant private persons are excluded.
Verified statements from source documents
Known links awaiting a source
Relevant case-law principles
ONE PERSON'S LIVING MEMORY
CannaInsider.eu — 1994 · 2004 · 2008 · 2010 · 2026
From an archive to an auditable decision graph.
Three publicly auditable evidence axes
Scale of the documented map
Active court branches in 2026
Submission and referral tree since 20 April 2026
Timeline and links
Candidate inconsistencies for verification
A CONCRETE USE CASE TODAY
A 2004–2026 European case as a network of institutions, proceedings, and remedies
The prototype displays an anonymized connection from the research program to the President, Czech courts and public prosecutors, ministries, police and forensic bodies, the CJEU, European Commission, and ECtHR. It distinguishes indexed material from creator-stated claims that still require an anonymized primary source.
Privacy in this demo: selected files are processed only in your browser. They are not uploaded and disappear when the page is refreshed.
EVERYDAY POTENTIAL
A person brings a box of records — the system returns an understandable path
A prisoner’s family, a disabled person, a senior, a debtor, or a parent dealing with an authority may fail not because evidence is absent, but because they cannot find it in time, connect it to the right proceeding, and explain it to a professional.
documents or speech in the user’s language→safe anonymization→timeline and evidence traffic light→review by a lawyer or advice service
Outcome: deadlines, institutions, unanswered arguments, and cited passages, with the proposed next step remaining under human review. Voice input, OCR, automatic anonymization, and universal analysis are product vision; the current public prototype safely demonstrates the principle on one supported sample.
CONTROLLED PRE-SUBMISSION TEST
Citizen XY’s archive already exists. One new document is being added.
2004–2026 archive→new referral notice→deadline and relevance check→updated decision graph
This button runs the first internal test. A separate outside-PDF test and safe text mapping appear below.
INTERNAL TEST RESULT
What the document proves
What the document does not prove
Relevance traffic light
Impact on the three unfiled documents dated 18 July 2026
Second layer: ministers, the Supreme Public Prosecutor, and other authorities
Questions for human verification
REAL-WORLD PROJECT OUTPUT
Documents sent on 19 July 2026
Created within AI Advocate for the Poor through collaboration between Mgr. Dušan Dvořák and Codex. At the author’s request, each document is published in its sent form, including his identifying details. The analytical layer identifies other people only by public role; the linked PDFs preserve the form in which the documents were actually sent.
BEFORE THE TEST
AFTER THE TEST
Outputs sent on 19 July 2026
The system created a new procedural node, assigned document-specific relevance, limited the meaning of the referral, and corrected document status. The three sent outputs appear below.
SECOND TEST — PUBLIC INPUT
A safe outside-input test — the PDF stays on your computer
Files are processed only in the browser and are never uploaded. The exactly supported court PDF is recognized by its digital fingerprint and produces an anonymized supervisory-board change analysis. An unknown PDF is safely rejected. Other text receives only a technical metadata map without inheriting prepared legal conclusions.
Open the anonymized evidentiary extract PDFOUTSIDE-PDF TEST RESULT