JURIS Architecture and Swiss Justice Data Sources
This document summarizes the Swiss justice architecture around JAA, justitia.swiss, and GoPro/Casedoc, and it identifies the data sources most suitable for building thalabus-based semantic similarity search.
Summary
Data sources
- Swiss Case Law Dataset (~1M cases): primary semantic corpus. It aggregates Swiss judicial decisions from multiple courts, including cantonal and federal courts. It is typically compiled for legal-AI research and derived from official court publication portals. It is not a single official dataset, and it may exist in multiple versions depending on the research group.
- Swiss Court Decisions (SCD) dataset (~127k cases): structured metadata dataset derived from Federal Supreme Court decisions. It is useful for metadata enrichment such as legal area, chamber, and decision type, but it is not a complete jurisprudence corpus.
- Swiss Federal Supreme Court decisions: original judgments published by the Federal Supreme Court, including ATF and non-ATF decisions. These represent the most authoritative jurisprudence layer.
- Court portals: official publication portals of federal and cantonal courts. These are the best source for incremental ingestion of newly published decisions.
- OpenCaseLaw: an open legal-data initiative containing case law from multiple jurisdictions. It may include Swiss decisions, but it should be treated as a supplementary source rather than the main corpus.
The JAA repository and the justitia.swiss platform are not historical jurisprudence databases and are not intended for precedent search. They function as secure judicial document exchange and electronic case-file infrastructure.
Key AI possible features
- Document intelligence
- Summarizing filings
- Extracting facts
- Evidence analysis
- Judicial drafting assistance
- Generating draft rulings
- Legal research
- Cross-case similarity search
The indexed corpus should support semantic similarity search combined with metadata filtering by court, date, and legal domain.
Prototype scope
- Prototype corpus: ingest about 10% of the Swiss Case Law Dataset, or roughly 100,000 cases, to validate semantic similarity search performance.
- With the current category-semantics algorithms, ingestion may take 2 to 3 weeks on t2b hardware, and less time if cloud resources are used.
- Prototype AI feature focus: legal research through cross-case similarity search.
Production scope
Minimal viable corpus
- Swiss Case Law Dataset as the multi-court base corpus
- Federal Supreme Court decisions as the authority layer
Production-grade corpus
- Swiss Case Law Dataset as the full base layer
- Federal Supreme Court decisions as the authority layer
- Federal Administrative Court decisions for administrative-law depth
- Continuous ingestion of newly published decisions from court portals
1. JAA in the Swiss digital justice stack
JAA, or Justizakten-Applikation, is an electronic case-file system used by Swiss courts and prosecutors. It is part of the Justitia 4.0 program.
It is not a public case-law repository. It is the operational digital file for a case.
Main functions
- Digital case-file management
- Secure storage of documents and evidence
- PDF annotation and review
- Task and workflow management
- Search across case documents
- Collaboration among judges, clerks, and prosecutors
The electronic file becomes the leading file for the proceeding.
Relationship to other systems
| Component | Role |
|---|---|
| justitia.swiss | Secure national platform for exchanging judicial documents |
| JAA | Internal case-file system used by courts |
Typical flow:
Lawyer / authority
-> justitia.swiss (secure submission)
-> Court IT systems
-> JAA (internal digital case file)
-> Judges / clerks work on the case
Operational context
- The platform is intended for roughly 300 Swiss courts and prosecutors.
- It is intended for roughly 30,000 justice-system professionals.
- It was adapted from an Austrian judicial workstation rather than built from scratch.
- The hosting model is SaaS with tenant separation by cantonal justice authority.
Security model
Secure connections are typically described as relying on:
- Site-to-site VPN
- Firewall rules
- Certificate authentication
2. GoPro and Casedoc
The GoPro referenced in Swiss justice documentation is a European enterprise software vendor, not the camera manufacturer.
What it provides
The relevant software family includes case-management and document-management products such as Casedoc and GoPro Foris.
Typical capabilities:
- Case lifecycle management
- Judicial document storage
- Workflow automation
- Secure records management
- Integration with government systems
These systems are used by courts, public administrations, banks, and regulators.
Where GoPro fits in the architecture
GoPro/Casedoc is best understood as the operational court layer, not the national exchange layer and not the jurisprudence repository.
Lawyers / Prosecutors
-> justitia.swiss
-> Court case-management system (for example Casedoc / GoPro-based solution)
-> Digital case repository (JAA or equivalent)
In this model:
- Casedoc = workflow and task orchestration
- JAA = file archive / court dossier repository
- justitia.swiss = secure transport layer
3. What this means for thalabus integration
The most realistic integration point for thalabus is the case-management layer, not JAA directly.
JAA is usually described as:
- Heavily secured
- Limited in API exposure
- An internal repository
- A legal archive
That makes it a poor place to implement semantic similarity search directly.
Where thalabus can fit
thalabus can act as an external AI reasoning and retrieval layer next to the case-management system.
Judge / Clerk
-> Case Management System (Casedoc / GoPro)
-> thalabus AI services
-> JAA
Potential AI tasks:
- Summarizing filings
- Extracting facts
- Generating draft rulings
- Evidence analysis
- Legal research
- Cross-case similarity search
Practical integration points
- REST APIs from Casedoc
- Document events for new filings or evidence
- Metadata hooks
- Document export/import pipelines
4. Why similarity search is not exposed by justitia.swiss
The API documentation for justitia.swiss describes a document transport and mailbox workflow, not a searchable repository of judgments.
What the API supports
- Upload of legal submissions
- Retrieval of documents sent to a party
- Access-right management
- Delivery receipts and hashes
- Document download
What it does not expose
- Court-case search
- Judgment search
- Similar-decision search
- Public jurisprudence queries
In other words, justitia.swiss is a transport layer, not a jurisprudence search engine.
Typical API model
Application (law firm software / court FA)
-> REST API
-> justitia.swiss
-> Mailbox / Sendings
-> Files (PDF, DOCX, etc.)
Conceptually, the API exposes messaging operations such as:
- createSending()
- listMailbox()
- downloadDocument()
- grantAccess()
It does not expose cases, judgments, or legal metadata as a searchable corpus.
5. Why GoPro should not rely on JAA for precedent search
If GoPro or Casedoc needs to find similar past judgments, there are three possible source categories:
- Public jurisprudence databases
- Local court archives
- JAA documents, theoretically, but with major confidentiality constraints
The JAA option is the least practical because documents are dossier-specific and access rights are case-specific. A global semantic search across all case documents would be extremely restricted.
6. Recommended architecture for thalabus
The most practical architecture is to build an external precedent index from public Swiss jurisprudence data.
Swiss case-law datasets
-> data normalization
-> AI embeddings
-> vector database
-> thalabus AI reasoning layer
-> GoPro UI
What GoPro would query
- Find similar past judgments
- Return relevant precedents
- Return reasoning excerpts
- Return similarity explanations
Suggested REST interface
- POST /legal/similarity
- Input: case description, legal articles, jurisdiction
- Output: ranked list of precedents
- POST /legal/explain_precedent
- Output: reasoning summary and extracted legal principle
7. How Swiss courts evaluate similar cases
Swiss courts do not determine similar cases purely by textual resemblance. A useful retrieval system should model similarity along multiple dimensions.
Core dimensions
| Dimension | Meaning | Example |
|---|---|---|
| Factual constellation | Real-world situation | Asylum claim based on desertion |
| Legal norm applied | Statutes or treaty articles | Art. 3 ECHR |
| Procedural posture | Appeal, first instance, revision | Asylum appeal |
| Jurisprudential authority | Hierarchical weight of the decision | Federal Supreme Court vs canton |
Authority hierarchy
| Rank | Court |
|---|---|
| 1 | Swiss Federal Supreme Court |
| 2 | Swiss Federal Administrative Court |
| 3 | Cantonal courts |
| 4 | Foreign jurisprudence |
| 5 | Academic doctrine |
Example scoring model
similarity_score =
0.45 semantic similarity
+ 0.25 legal article overlap
+ 0.15 precedent citation proximity
+ 0.10 court hierarchy weight
+ 0.05 temporal relevance
8. Graph and vector architecture
The most effective system combines embeddings, metadata filters, and a citation graph.
Processing pipeline
Court decision
-> document parser
-> fact extraction
-> legal-reference extraction
-> citation extraction
-> vector DB
-> graph DB
-> metadata index
Storage layers
| Storage | Purpose |
|---|---|
| Vector database | Semantic similarity |
| Neo4j | Citation network |
| Relational index | Metadata filtering |
Example graph model
Nodes:
- Case
- Article
- Country
- Court
Relationships:
- Case cites Case
- Case applies Article
- Case originates from Country
- Case decided by Court
Example query patterns
MATCH (c:Case)-[:APPLIES]->(a:Article {id: "Art3_ECHR"})
RETURN c
MATCH (c:Case)-[:CITES]->(p:Case)
WHERE p.id = "BVGE-2015-3"
RETURN c
9. Practical conclusion
If GoPro wants thalabus to provide AI similarity search, the best route is:
- Build a corpus from public Swiss jurisprudence datasets.
- Normalize and anonymize the documents.
- Index them in a vector database and a citation graph.
- Expose thalabus as an external AI similarity API.
- Integrate that API into GoPro/Casedoc at the case-management layer.
That approach keeps JAA as the official case-file system while using thalabus for search, ranking, and reasoning.
10. Sources and references
- https://www.justitia.swiss/downloads/justitia.swiss/api-justitia.swiss-for-developers-v49-20250711_111343.pdf
- https://www.justitia.swiss/de/entwickler/applikationsdokumentation