Semantic Legal Search Platform
Semantic search
RAG
On-Premise
LLM
Legal
IA
Elasticsearch
“OSMOZ-IT & MATICOM were able to understand our confidentiality constraints and offer us a modern AI solution without compromising on the security of our data. Hybrid search will transform the way we work.”
Contexte
The challenge: FBT, a Swiss law firm, was faced with a growing volume of documents (case law, legal doctrine, legal texts, contracts), making keyword searches time-consuming. Identified needs :
- Intelligent search with specialized legal vocabulary
- Absolute confidentiality of customer data (on-premise hosting mandatory)
- Traceability of sources cited for each result
Solutions
We designed and deployed a 100% on-premise RAG (Retrieval-Augmented Generation) platform, enabling intelligent semantic search across the entire documentary corpus. Architecture implemented :
- Vector search engine: Elasticsearch with 768-dimension embeddings
- Local LLM: Ollama (Mistral) for contextualized response generation
- Embedding template: nomic-embed-text optimized for legal French
- Secure API: Spring Boot with OAuth2/Keycloak authentication
- Dedicated infrastructure: High-performance server (AMD Ryzen 9, 30 GB RAM allocated)
Key features :
- Search by semantic similarity (not just keywords)
- Automatic indexing of new documents
Résultats
Response time: < 10 seconds Confidentiality: 100% on-premise, no data outsourced Relevance: Hybrid semantic & lexical search (keywords)
NOUS CONTACTER
Vous avez un projet similaire ?
Vous souhaitez exploiter l'IA pour valoriser vos documents tout en gardant le contrôle total de vos
données ?
