← Back to Technology DEEP TECH ARCHITECTURE

SemanAI Hybrid RAG Architecture v1.0

Neuro-Symbolic Engine & Data Pipeline Specification

Data Ingestion πŸ›οΈ Public Datasets - AI Hub (Healthcare) - K-Data (Senior Log) πŸ“‚ Internal Logs - App User Activity ETL & Processing Unstructured Pipe 1. Chunking 2. Embedding Model Model: bge-m3 (Kor) Structured Pipe 1. Entity Extraction 2. Relation Mapping Tech: RE-Alpaca Vector DB Milvus / Chroma Graph DB Neo4j / Memgraph (Ontology-aligned KG) SemanAI GraphRAG Engine 1. Hybrid Retrieval Vector Sim. (kNN) + Graph Traversal (Cypher) -> Re-Ranking 2. Generation (LLM) Context Injection: "Patient A (Parkinson) + Floor(Tile)" Model: Llama-3 (Frozen Weights) Instruction-Tuned Base 3. Fact Check (Guardrail) Factual Consistency Check via Graph 4. Decision-Level Reasoning Policy & Risk Rules (Approve / Escalate) Safe Prediction

이 λ³΅μž‘ν•œ 과정을 λˆ„κ°€ ν†΅μ œν•˜κ³  μ±…μž„μ§€λŠ”κ°€?

μ˜μ‚¬κ²°μ • κ±°λ²„λ„ŒμŠ€ (Decision Protocol) βž”