Clément Dhollande French Translator and Tutor

Leurn App

Leurn App is a high-precision Computer-Aided Translation (CAT) environment. It is engineered for linguistic professionals who require the power of Neural Information Retrieval without the security risks of cloud-based AI.

• Primary Function: Advanced semantic memory and automated compliance auditing.
• Deployment: Standalone desktop application (No external dependencies or APIs).
• Security: Native offline operation; air-gap compatible.

Discover the demos here

The Neural Engine

The application utilizes a proprietary dual-model pipeline designed to handle complex linguistic tasks on consumer-grade hardware.

• Semantic Retriever: A multilingual transformer model that maps conceptual meaning across 50+ languages. Unlike keyword search, it identifies “intent” and “context” with high dimensional accuracy.
• Contextual Reader: A specialized extractive logic engine that identifies specific data points within large documents. It is non-generative, meaning it cannot “hallucinate” or create false data.
• Hardware Optimization: * Native acceleration for Apple Silicon (M-series) and NVIDIA (CUDA).
  • Optimized CPU fallback for legacy hardware (4GB RAM minimum).

Core Capabilities

A. Neural Workbench (Translation Memory)

• Vector-Based Indexing: Converts legacy TMX, Excel, and Word files into high-speed semantic stores.
• Intelligent Highlighting: Automatically identifies and renders the exact phrase matches within a source/target pair for instant verification.
• Hybrid Search: Supports Semantic (Concept), Exact (Regex), and Fuzzy matching.

B. Knowledge Grid (Advanced RAG)

• Document Intelligence: A custom hierarchical parser that preserves the structure of complex PDF style guides and policy documents.
• The “Zero-Risk” Pipeline: Uses Retrieval-Augmented Generation (RAG) principles to find “ground-truth” answers from your own documents without sending data to a third-party LLM.

C. Neural Alignment

• Meaning-Based Mapping: Aligns source and target files by comparing semantic distance, allowing for accurate alignment even when sentence counts differ between languages.

Data Architecture & Privacy

Leurn App is built on a “Zero-Leak” philosophy.

• Storage: All data is stored in a flat-file, portable structure within the local workspace.
• Data Sovereignty: No data ever leaves the local machine. There is no telemetry, no cloud syncing, and no external API calls.
• Interoperability: Full support for industry-standard formats including .xliff, .tmx, .docx, and .xlsx.

Performance Benchmarks

(Tested on standard M1 hardware)

• Startup: < 4 seconds.
• Search Latency: Sub-50ms (Instantaneous retrieval).
• Index Speed: High-speed processing of ~20 pages per second.
• Resource Efficiency: Active memory footprint remains under 1GB, even during complex neural tasks.