The JuDDGES Ecosystem
A collaboration, not a monolith
JuDDGES is intentionally built as a federation of focused tools rather than a single application. Each project is independently deployable, has its own maintainers, and lives in a separate Git repository — but they share a common mission and form a coherent research pipeline. This page maps out who builds what, and how the pieces fit together.
The work is shared between the team at Wrocław University of Science and Technology (this platform and the parent research project) and partner contributors maintaining the human-in-the-loop annotation workbench. Both sides release their code under permissive open-source licenses, so the entire pipeline — from raw judgment text to verified annotations to user-facing search — can be inspected, reproduced, or extended by anyone.
Projects in the ecosystem
Each repository is open-source and independently versioned.
JuDDGES (research)
Parent projectParent research project
Maintained by: Wrocław University of Science and Technology
The umbrella research initiative on legal NLP and human-in-the-loop machine learning for Polish and England & Wales judicial decisions. Hosts datasets, NLP pipelines, model training code, and the published Hugging Face datasets that power downstream tools.
HITL Annotation Tool
UpstreamUpstream — human annotation workbench
Maintained by: Middlesex University, London (UK partner team)
A dedicated human-in-the-loop annotation workbench used by domain experts to verify and correct LLM extractions over UK criminal court judgments across 43 canonical legal fields. Produces verified ground-truth annotations that feed back into JuDDGES datasets and benchmarks.
JuDDGES App
You are hereThis platform — research instrument
Maintained by: Wrocław University of Science and Technology
The web platform you are using right now. Provides hybrid semantic + full-text search, dynamic schema-driven extraction, and analytics over Polish and England & Wales judgments. Designed as an open research instrument that consumes datasets and models produced upstream.
How data flows through the pipeline
From raw judgment text to research insight, in three stages — each owned by a different part of the collaboration.
Human annotators verify
Legal experts review and correct LLM-generated extractions in the HITL Annotation Tool, producing high-quality ground-truth spans for 43 canonical fields.
HITL Annotation ToolDatasets are published
Verified annotations and curated judgment corpora are released on Hugging Face under the JuDDGES organization, openly available to the research community.
huggingface.co/JuDDGESResearch is enabled
Researchers query and extract structured data from judgments through this platform — backed by the datasets and models produced upstream by the wider JuDDGES collaboration.
JuDDGES App (this site)Why a federated ecosystem?
Annotation workbenches and research platforms are very different beasts. Annotation tools optimize for expert-user ergonomics, low-friction span editing, and dataset export. Research platforms optimize for scale, hybrid retrieval, multi-tenant access control, and rich analytics. Mashing them into one codebase would force compromises on both sides.
By keeping the projects separate but linked through shared open data formats (Hugging Face datasets, JSONL exports, common code lists), each team can iterate on their own cadence, target their own users, and still contribute to a coherent whole.
Shared standards
Verbatim span extractions, controlled vocabularies of legal codes, and the Hugging Face dataset format are the lingua franca between projects.
Independent releases
Each repo has its own versioning, CI, and deployment pipeline. Breaking changes in one tool don't block work in the other.
Open by default
All code is published under permissive licenses; datasets are released openly on Hugging Face for reproducibility and reuse.
Who maintains what
The collaboration is split between two academic teams across two universities.
Explore the ecosystem
Direct links to the repositories, project website, and shared resources.
Want to collaborate?
Both repositories welcome contributions. Open an issue, propose a feature, or join the discussion.