// CASE STUDY
Rosendahl — Legal AI Platform (LunarTech)
Engineered backend services and Agentic RAG pipelines for Rosendahl, LunarTech's self-hosted legal AI platform — a 'Claude Code for law' operating environment that structures ECHR legal cases like codebases. Built legal document processing APIs (LexNLP, Blackstone/spaCy) for named-entity extraction, citation parsing, and admissibility-risk scoring. Enforced a four-tier confidential AI model routing strategy (OpenRouter / Claude / Ollama local) across 121+ active case workspaces.
- Role
- Backend AI Engineer
- Year
- 2026
- Context
- LunarTech
- Status
- In production
Overview
Rosendahl is an enterprise-grade legal AI platform built for LunarTech. It acts as an intelligent operating environment that treats complex ECHR legal cases like codebases, providing attorneys with an autonomous 'Claude Code for law' experience.
The problem
Legal professionals spend countless hours manually parsing unstructured case files, searching for citations, and assessing admissibility risks. Existing RAG solutions lack the domain-specific rigor and strict data-confidentiality controls required by law firms.
My approach
I engineered advanced Agentic RAG pipelines and custom legal document processing APIs using LexNLP and Blackstone/spaCy. These services automatically extract named entities, map complex citation graphs, and compute admissibility-risk scores across massive document corpora.
Architecture decisions
Confidential model routing — Implemented a robust four-tier routing strategy (OpenRouter, Claude, and local Ollama deployments) to ensure highly sensitive legal data never leaves the secure boundary.
Agentic orchestration — Utilized LangGraph and LlamaIndex to build stateful AI agents capable of multi-step reasoning and deep contextual retrieval across 121+ active case workspaces.
Outcome
Rosendahl radically accelerated case preparation timelines. The platform successfully automates the extraction and structuring of dense legal arguments while maintaining strict compliance with legal data-privacy standards.