Knowledge Base
Understanding Knowledge Systems in Lyzr
A comprehensive overview of Lyzr’s three core knowledge systems—Knowledge Base, Knowledge Graph, and Semantic Model—to help you choose the right tool for your agents and data.
Lyzr provides three powerful frameworks to structure, retrieve, and reason over your data:
Knowledge Base, Knowledge Graph, and Semantic Model. Each serves a specific purpose in enabling intelligent AI behavior for different kinds of information. This guide will help you understand what each system does, when to use it, and how they complement each other.
Knowledge Base, Knowledge Graph, and Semantic Model. Each serves a specific purpose in enabling intelligent AI behavior for different kinds of information. This guide will help you understand what each system does, when to use it, and how they complement each other.
🔍 Knowledge Base (KB)
The Knowledge Base is a retrieval-first system designed for unstructured content like PDFs, DOCX, TXT, or web pages. It breaks your documents into chunks, embeds them as vectors, and retrieves the most relevant pieces for LLMs to answer user queries.Best for:
- Domain-specific documents
- FAQs, manuals, policy documents
- Basic RAG (Retrieval-Augmented Generation) pipelines
How it works:
- Chunking: Splits documents into context-rich sections.
- Vectorization: Embeds chunks into a vector store (e.g., Pinecone, Qdrant).
- Retrieval: Finds top-matching chunks when a question is asked.
- Generation: LLM creates answers using retrieved content.
Benefits:
- Quick setup with minimal preprocessing
- Flexible retrieval strategies: Basic, MMR, HyDE
- Works well with general text sources
🌐 Knowledge Graph
The Knowledge Graph transforms your unstructured data into structured, interconnected nodes and relationships, allowing for advanced multi-hop reasoning and deeper understanding. Built on Neo4J, it excels in scenarios where relationships between concepts matter as much as the content itself.Best for:
- Complex interlinked content
- Entity-centric domains (e.g., legal, healthcare, HR)
- Cross-document or multi-hop questions
How it works:
- NER & Relation Extraction: Identifies entities and their links.
- Graph Construction: Builds a Neo4J graph with nodes & edges.
- Graph Querying: Enables traversal-based retrieval for LLMs.
Benefits:
- High accuracy on relational queries
- Low hallucination rate for complex reasoning
- Visual exploration and traversal possible
📊 Semantic Model
The Semantic Model bridges the gap between structured tabular data (like SQL tables, CSVs) and natural language understanding. It enriches tables with human-readable descriptions so AI agents can explain, search, and query data effectively.Best for:
- Internal tools and analytics teams
- AI assistants for data documentation
- Natural language interfaces over databases
How it works:
- Column-level Descriptions: Auto-generates purpose for each column.
- Table Summaries: Explains what each table is about.
- Vector Embedding: Enables semantic search and natural language querying.
Benefits:
- Makes raw tables understandable by humans and AI
- Powers doc agents, RAG, and search over databases
- No need for deep technical understanding of the schema
🧠 Choosing the Right Knowledge Tool
| Use Case | Use this system |
|---|---|
| General unstructured documents | Knowledge Base (KB) |
| Questions about complex relationships | Knowledge Graph |
| Tabular data with structure | Semantic Model |
| Entity-specific reasoning | Knowledge Graph |
| Policy or FAQ automation | Knowledge Base |
| Internal DB or table QA | Semantic Model |
| Multi-hop or “how does X relate to Y” | Knowledge Graph |
| Explaining column meanings | Semantic Model |
🧩 Combined Usage (Hybrid AI)
These systems are not mutually exclusive. In many cases, combining them leads to better AI performance:- Use Knowledge Base for broad document access.
- Overlay a Knowledge Graph for deep relationship reasoning.
- Attach a Semantic Model for querying your structured data.
Ready to Build?
Start by understanding your data type and the problem you’re solving:- If it’s documents, go with the Knowledge Base.
- If it’s about connections, use the Knowledge Graph.
- If it’s tabular or relational, build a Semantic Model.