AdmissusCase: Legal AI for Human-Rights Casework

Casework context

Preparing material for the European Court of Human Rights or UN treaty bodies combines large case files, timelines, deadlines, and procedural requirements. In a small team, document organisation competes directly with time available for legal analysis.

AdmissusCase is a product hypothesis for this workflow: the system should prepare an organised dossier without replacing the lawyer's judgment, accountability, or signature.

Problem

Automation without a clear scope can create false confidence in completeness. A fluent summary does not reveal a missing attachment, inconsistent date, or passage without provenance. In high-stakes work, omissions must remain visible.

My role

• Co-Founder and Technical Lead, as stated on the public landing page;
• co-development of the product hypothesis;
• architecture of the technical document flow;
• evaluation-plan and accountability-boundary design.

Product flow

A private-repository review confirmed that the prototype separates document upload and storage, dispatch to an external processing service, callback handling, task status, and credit history. The dashboard gives users a view of documents and results without collapsing every layer into one opaque step. These details describe an internal code review and are not presented as a public artifact.

document upload → storage → task dispatch → processing callback → result history → lawyer review

This separation helps identify where a failure occurred and whether a result belongs to the correct file and task version.

The system prepares, the lawyer decides

The system may organise material, assist extraction, construct a timeline, and prepare a draft for review. The lawyer assesses file completeness, legal qualification, factual significance, and readiness for use. This boundary is part of the product rather than a disclaimer added afterwards.

How quality should be measured

The benchmark plan covers dossier completeness, citation fidelity, document-to-fact links, and corrections requiring a specialist. Results should use explicit test cases and methodology rather than an attractive percentage alone.

What was achieved

The work produced a public presentation of the product direction. An internal code review confirmed an architecture spanning frontend, backend, storage, asynchronous tasks, and credit history. The case demonstrates product and technical responsibility in social-impact LegalTech.

Evidence we do not use

Template testimonials and claims about time savings, near-zero hallucinations, grants, or certifications are not carried over without documentation and methodology. A public benchmark remains a planned artifact.

Limitations

Current product maturity, grants, and certifications require owner confirmation. This material is not legal advice, and the prototype does not make final decisions on the lawyer's behalf.