📥
1. Ingest
Upload PDF or markdown and enqueue indexing through the API or console.
Architecture
TurboDex keeps document hierarchy intact so your AI answers with context, not fragments.
Instead of flattening documents into disconnected chunks, TurboDex builds a navigable tree with section lineage, summaries, and query-ready metadata.
The Four-Stage Pipeline
Every document uploaded to TurboDex flows through four deterministic stages.
📊
2. Parse
TurboDex identifies TOC structure, page boundaries, and hierarchical relationships.
📌
3. Persist
The tree is persisted in YottaDB with node-level metadata and retrieval lineage.
🔍
4. Retrieve
Queries traverse hierarchy-aware nodes for better answer grounding and summaries.
Traditional RAG Pipeline vs TurboDex Pipeline
✗ Traditional Chunk / Vector Flow
Document
→
Chunking
→
Embeddings
Vector Search
→
Candidate Snippets
→
Answer
Section lineage is lost at the chunking step. Retrieval can mix semantically similar but structurally unrelated content — a risk factor paragraph alongside a financial projection paragraph looks identical to a vector index.
✓ TurboDex Hierarchical Flow
Document
→
Structural Parsing
→
Hierarchy Tree
Node Summaries
→
Context Traversal
→
Answer
Retrieval respects document anatomy. The engine knows that "Section 4.2.1 Risk Factors" is a child of "Section 4 Management Analysis" and answers accordingly. No structural drift.
What the Hierarchy Tree Captures
Each node in the TurboDex tree stores structured metadata beyond plain text.
Section Lineage
Full path from root to leaf — Root › Chapter › Section › Paragraph — preserved per node.
Node Summaries
AI-generated summaries at each structural level, enabling hierarchical answer synthesis.
Page & Token Metadata
Physical page ranges, token counts, OCR flags, and document boundary markers per node.
YottaDB Persistence
Stored in a high-performance hierarchical key-value store. Sub-millisecond node lookup without re-embedding.
Query-Ready Metadata
Every node is indexed and traversable by title, depth, parent, child, and semantic distance.
OCR & Scanned PDF Support
Automatic OCR detection and fallback for image-only pages. OCR page count tracked per document.
Ready to test with your own document set?
Start in SaaS for immediate access, or model private-cloud economics if compliance and data control are priorities.