Lex AI - reasoning agent for Serbian legal research
A domain-specialized reasoning agent operating natively inside Paragraf Lex, the de facto legal information platform in Serbia. In production since 2026, deployed under continuous evaluation.
AI built around the data, not the model.
We build production AI for organizations whose work depends on regulated content, complex data, and high-stakes reasoning.
- SOLUTIONS IN PRODUCTION
One product. Two engagements. Live in 2026.
Lex AI is our own product, deployed commercially through Paragraf Lex. Two active engagements for the Art Loss Register are going live in 2026.
Catalog Star
An engagement for the Art Loss Register - structuring the historical auction record for due-diligence search. In phased rollout; going live in 2026.
Classic Car Register
An engagement for the Art Loss Register - linking visual evidence to provenance claims across decades and auction houses. In phased rollout; going live in 2026.
- HOW WE WORK
Engineering, not prompting.
Our process treats AI systems as engineering problems. Retrieval, structure, evaluation and integration come before model choice - because models change every year, but the data and the domain don't.
01
Domain modelling
We co-develop the data ontology with subject-matter experts before any model is selected. Domains carry constraints models alone won't surface.
02
Retrieval & extraction architecture
Hybrid retrieval, structured parsing and entity linking are designed up front. The retrieval layer determines what the model can know.
03
Reasoning layer integration
Models - closed-source, open-weights or fine-tuned - are selected to fit the reasoning load, not the other way around.
04
Evaluation harness
We build the eval before the system. Golden sets, retrieval metrics, hallucination measurement, regression tests.
05
Production integration
Our systems ship inside existing platforms with auth, audit, monitoring and an operational handover.
06
Continuous evaluation
Production traffic feeds back into the eval. Quality is measured on a recurring cadence, not at launch and never again.
- WHY PROZONE AI
Built on enterprise software experience, focused on modern AI.
Prozone AI is the AI-native continuation of the Prozone software ecosystem - decades of enterprise delivery experience now pointed at the systems organizations need to build with modern machine learning.
01
Enterprise heritage
Built on years of delivering complex software systems, document management and business-process platforms across regulated sectors.
02
Domain depth, not generic chat
We focus on AI systems that genuinely understand legal, archival and enterprise data - not generic LLM wrappers.
03
Engineering across the stack
Retrieval, structured extraction, agentic workflows, predictive ML and evaluation - delivered as production software with operational ownership.
04
Regional reach
Operating through Prozone AI and Prozone Middle East, with delivery capability across Serbia, Europe and the Gulf region.
- TRUST & DEPLOYMENT POSTURE
Where the system runs is part of the design.
Engineered for organizations where data sensitivity, audit obligations and operational accountability are primary - not afterthoughts.
Your data, your infra
On-prem · Private cloud · Hybrid
Audit-ready
Logging · Evaluation · Regression testing
No data exfiltration
Conservative on what leaves your network
Production accountability
Operational ownership, not one-shot delivery
Have a domain that nobody else seems to understand?
That's usually where we start. Tell us about the documents, the data, the constraints - and what your AI system needs to actually understand.