Oxaide Knowledge
The Enterprise Reasoning Engine. Sovereign RAG for critical infrastructure, field ops, and institutional memory.
The Pipeline Architecture
1. Extraction
OCR & Text Parsing
2. Chunking
Semantic Splitting
3. Embedding
Vector Generation
4. Indexing
pgvector Storage
Chunking Strategy
Context PreservationWe don't just split by character count. Oxaide uses Recursive Semantic Chunking to keep related paragraphs together. This ensures that when a "Rule" is cited, its "Exceptions" are included in the same context window.
Retrieval (RAG)
Hybrid SearchDense Vector Search
Understanding "Concepts" (e.g., matching "invoice" with "bill"). Powered by OpenAI text-embedding-3-small (Cloud) or nomic-embed-text (Sovereign).
Sparse Keyword Search
Matching exact "Terms" (e.g., "IM8-Clause-5.1"). Essential for regulatory compliance where exact wording matters.
Deployment Models
Flexible InfrastructureCloud
Standard SaaS deployment. Hosted on AWS Singapore. Best for speed and scalability.
Airgap
Fully offline. No internet access required. Runs on local hardware for maximum security contexts.
Sovereign
Private Cloud or On-Premise. Full data sovereignty with nomic-embed-text for local embeddings.