Skip to main content
The Knowledge layer is an entity‑centric model that organizes matter‑specific facts to support accurate retrieval and contextual reasoning across large document collections. It provides a persistent foundation for Epiq AI by structuring information in a way that both users and AI systems can reason over consistently. The Knowledge layer processes documents and extracts knowledge from both text and metadata. It organizes that knowledge into structured entities, relationships, topics, and supporting metadata so information can be queried, connected, and reused across a matter without relying on isolated document analysis.

​
Core components

The Knowledge Layer is built on four primary components.

​
Entities

An entity is a structured representation of a person, organization, topic, or event extracted from source documents. Each entity includes:
  • Structured attributes (for example, names, roles, or affiliations)
  • Citations that link attributes back to source documents
By organizing information around entities, the Knowledge layer establishes contextual connections across a matter. This allows systems to understand relationships among people, organizations, topics, events, and evidence at a factual level.

​
Protocols

Protocols are defined sets of instructions, rules, or structured content associated with a matter. Protocols are indexed and embedded into the Knowledge layer so they can inform retrieval‑augmented generation (RAG) processes. The Knowledge layer stores protocol‑related information such as:
  • Protocol summaries
  • Session breakdowns
  • Extracted entities
  • Additional protocol metadata or structured details
Topic entities related to a protocol are explicitly linked. This allows retrieval and reasoning to incorporate both document‑derived facts and protocol‑defined guidance.

​
Metadata

Metadata provides descriptive context that supports reasoning and retrieval. When documents are supplied to large language models, selected metadata fields are included to enhance understanding. Examples of metadata include:
  • Authors
  • Custodians
  • Reviewer‑applied tags
  • Entity attributes
Metadata combines automated extraction with human‑supplied insights. Depending on the query and retrieval logic, related entities and their attributes may also be included as part of the context provided to the model.

​
Temporal modeling

The Knowledge Layer supports temporal accuracy when people or organizations change over time. Key behaviors include:
  • Capturing start and end dates for roles or titles when available
  • Consolidating information into a single profile when dates cannot be determined
  • Modeling employment and affiliation as person‑organization pairs
  • Creating separate entries when a person changes organizations or when distinct employment periods are identified
This approach preserves historical context while clearly representing role evolution.

​
How knowledge is stored

Extracted knowledge is stored across multiple database types to support different access patterns:
  • Relational database. Stores entities, relationships, and metadata.
  • Text search database. Indexes textual topics and summaries.
  • Vector database. Stores vectorized representations (embeddings) for advanced search and clustering.
This layered storage approach enables both precise lookup and semantic retrieval.

​
Editing and management

You can edit entities and relationships within the Knowledge layer as needed. This flexibility allows refinement of extracted knowledge to better reflect the facts of a matter.

​
Summary

The Knowledge layer combines entity‑centric modeling, protocol awareness, structured metadata, and temporal context into a single system. Together, these elements enable accurate retrieval, richer reasoning, and deeper factual understanding across large and complex datasets within Epiq AI.