TECHNOLOGY

Ask one question across meaning and facts.

Hybrid RAG can search passages by meaning, follow exact relationships in a knowledge graph, and return the source material used for an answer.

TALK THROUGH A WORKFLOW COMPARE THE PATHS

VECTOR + GRAPH

Why does one question need two kinds of retrieval?

Select each stage to see where meaning, relationships, and citations contribute.

Skip the guided view

01File

Preserve the structure of the source.

A contract PDF becomes headings, clauses, tables, and page references instead of one undifferentiated block of text.

Preserved: Section names, page numbers, parties

ResultSource-aware chunks ready for indexing.

ILLUSTRATIVE TECHNICAL PATTERN

02Vector

Use meaning to find likely passages.

Vector search handles fuzzy language. A question about ending an agreement can find clauses titled termination, cancellation, or notice.

Best for: Similarity, themes, broad recall
Limit: A nearby passage is not automatically a verified fact

ResultA ranked set of semantically relevant passages.

03Graph

Use relationships to test exact facts.

The graph connects Vendor X to its active agreement, amendment, dates, obligations, and governing clauses.

Best for: Entities, relationships, exact filters
Limit: Accuracy depends on extraction and upkeep

ResultA fact path that can confirm which document governs.

04Hybrid

Combine recall with a precise relationship path.

Vector candidates and graph relationships are used together, then only the evidence relevant to this question enters the briefing.

Vector: Candidate clauses
Graph: Active agreement and amendment

ResultA smaller evidence set with fewer unresolved conflicts.

05Cite

Make the retrieval path inspectable.

The answer links its claim to the source clause and identifies uncertainty when the available records do not resolve the question.

Answer
Source file and section
Uncertainty flag when needed

ResultA response a reviewer can verify at the source.

Place retrieval inside a controlled workflow

Compare vector, graph, and hybrid retrieval in one view

VECTOR DATABASE

"Find everything similar to this." Great for fuzzy questions, summaries, and discovering related material. "What do our contracts say about early termination?"

KNOWLEDGE GRAPH

"Which entities connect to which?" Exact answers across relationships: customers, contracts, dates, obligations. "Which vendors have net-45 terms AND an active NDA?"

HYBRID RAG

Meaning for recall, facts for precision, with source references when the retrieval path supplies them. "Summarise our exposure if Vendor X terminates early."

The strongest retrieval design follows the questions people need to answer and the evidence those answers require.

Bring us a difficult knowledge question →