iuris.ba
Structured legal research, document analysis and evidence-grounded AI for Bosnia and Herzegovina.
iuris.ba is a legal-tech system built around structured legislation, court decisions, document workflows and retrieval-assisted analysis. The product separates legal data, application orchestration, vector retrieval and model-serving concerns so AI output can be grounded in traceable legal sources rather than treated as an isolated chatbot feature.
Product type
Legal-tech · research & analytics
Status
Controlled development
Astraealibra scope
Product architecture · legal data systems · retrieval/AI · full-stack development
Product identity
iuris.ba
Product in practice
An active domain with controlled access.
iuris.ba is connected to the real application system, but access is intentionally controlled while the legal corpus, document analysis and AI infrastructure are further hardened. The portfolio therefore reflects the actual product state rather than presenting development work as an unrestricted public service.
Domain
iuris.ba
Access
Controlled access
iuris.ba
https://iuris.ba
Functional scope
What the product is built to do.
Structured legislation and court-decision corpus for legal research
Normalization and linking of legal references across documents and jurisdictions
Semantic retrieval over legal material with vector search
Document upload and extraction workflows for analysis
Evidence-oriented AI analysis designed around citations and retrieved context
Controlled user access for production-stage legal tooling
Separated model, embedding, reranking and document-processing services
System architecture
How the product is structured beyond the interface.
Research surface
01A Next.js application provides legal research, law and decision views, analytics surfaces and controlled access through Clerk.
Legal application layer
02The Node.js backend remains responsible for authentication, orchestration, legal business rules, citations, logging and retrieval logic.
Data & retrieval
03Structured legal data is persisted in PostgreSQL while Qdrant supports semantic vector retrieval over normalized legal content.
AI runtime
04A separate Python/FastAPI document worker and OpenAI-compatible model-serving architecture support document extraction, LLM inference, embeddings and reranking without coupling model runtime to the application backend.
Engineering highlights
The work that makes the system durable, testable and operable.
Legal-reference normalization designed to avoid binding similarly named provisions to the wrong jurisdiction
Retrieval architecture that separates structured legal state from vector search and answer generation
Dedicated document-processing worker for PDF/DOCX/TXT extraction
Model-serving layer separated from the main backend so AI infrastructure can evolve independently
Testing and curated legal datasets used to protect parser and retrieval behavior from regressions
Technology
Production stack and supporting infrastructure.
Current direction
Where the product is now.
The platform is in controlled development while corpus quality, reference normalization, semantic retrieval, document analysis and model-serving architecture are hardened.