SOG Engine is a high-performance intelligence manifold designed for localized Retrieval-Augmented Generation (RAG). By transforming static technical documentation into a self-organizing high-dimensional graph, SOG enables real-time reasoning on edge hardware and high-end desktops without cloud dependency.
- Massive Document Sweeps: Engineered to ingest 160+ technical manuals in a single batch.
- Incremental Persistence: Implements "Sync-after-Doc" protocols to ensure data integrity during long ingestion cycles.
- Memory-Optimized: Specifically tuned for 32GB RAM environments, utilizing
ConcurrentHashMappre-allocation to prevent resizing overhead. - Hardware-Agnostic Inference: Dual-profile support for Ollama (Desktop) and Jetson Nano (Edge) via native Java providers.
- Topological Retrieval: Uses cosine similarity and keyword-synapse mapping to retrieve the most relevant context for LLM synthesis.
- Backend: Java 22, Spring Boot 3.x
- Inference: LangChain4j (Ollama & OpenAI-compatible drivers)
- Graph Logic: Self-Organizing Manifold (SOM) inspired vector space
- Hardware: Optimized for NVIDIA Jetson (ARM64) and Windows/Linux x86
- Java 22+ installed and configured.
- Ollama (Desktop) or llama.cpp (Jetson) running locally.
git clone [https://github.com/yourusername/sog-engine.git](https://github.com/yourusername/sog-engine.git)
cd sog-engine
mvn clean packagejava -Dspring.profiles.active=desktop -Xmx24G -jar target/sog-engine.jarUse the included tactical scripts for optimized edge performance:
chmod +x scripts/*.sh
./scripts/setup_jetson.sh
./scripts/deploy_sog.sh| Endpoint | Method | Description |
|---|---|---|
/api/sog/upload/batch |
POST |
Upload multiple manuals for manifold ingestion. |
/api/sog/query |
POST |
Query the SOG with natural language. |
/api/sog/ram-metrics |
GET |
Monitor real-time JVM memory utilization. |
The project includes a comprehensive, zero-warning JavaDoc suite. To generate the latest technical documentation:
mvn javadoc:javadocGenerated docs can be found in target/site/apidocs/index.html.
Distributed under the MIT License. See LICENSE for more information.
Developed for the next generation of localized, edge-based intelligence.