The Catalog Register
An engagement structuring millions of historical auction record for due-diligence search at industry scale.
Lex AI is our own product. We built it, and it operates natively inside Paragraf Lex - the de facto reference platform for Serbian legal professionals - through a commercial partnership. It compresses the path from a serious legal question to a sourced, verifiable answer.
Serbian legal language is technically dense and morphologically rich. Articles cross-reference each other in ways that don't survive naive search. Hallucinating in a legal product is worse than no answer.
Built around Paragraf Lex's editorial discipline. Source-anchored reasoning with explicit citations into the underlying material - the lawyer verifies in a click. We do not publish system internals.
Surviving production with a legal audience required deep domain modelling, sustained editorial partnership, and an evaluation posture that treats accuracy and provenance as primary metrics, not secondary ones.
Built around Paragraf Lex's editorial discipline. Their editors are co-authors of the product - their domain authority shapes what the system says, and what it refuses to say.
Both are in phased rollout and going live in 2026.
An engagement structuring millions of historical auction record for due-diligence search at industry scale.
A vehicle-aware retrieval engagement that links visual evidence to provenance claims across centuries and auction houses.
Methodology that applies beyond the engagements above - to regulated industries, financial services, the public sector, and industrial environments.
Source-anchored answers in domains where correctness is verifiable.
Layout-aware understanding for complex archives, contracts, claims, reports.
Hybrid lexical-semantic search with domain-tuned ranking; cross-modal where vision matters.
Visual evidence as first-class input - provenance, inspection, archival material.
Action-taking systems with bounded action surfaces and audit trails.
Forecasting, anomaly detection, decision-aware classification and ranking.
That's usually where we start. Tell us what you have, what it should do, and where the cost of being wrong lives.