KEYSTONE: Building an Evidence-First AI Company Brain with FalkorDB

Description and use

KEYSTONE uses FalkorDB to connect company data, detect conflicts, and provide evidence in answers, enhancing decision-making.

Problem

Company knowledge is connected but often scattered, making it hard to understand how different pieces of information relate and to detect conflicting claims.

Outcome

KEYSTONE demonstrates how a graph database can enable evidence-aware AI by connecting claims to projects, owners, and risks, detecting conflicts, and presenting evidence alongside answers, thereby improving decision-making.

Workflow

• 1. Ingest company documents and extract entities and claims.
• 2. Store entities, claims, and relationships in FalkorDB as a knowledge graph.
• 3. Visualize the knowledge overview in a Streamlit dashboard.
• 4. Detect potential conflicts by comparing related claims (e.g., launch dates).
• 5. Flag conflicts and encourage users to review underlying evidence.
• 6. Connect projects, owners, and risks to provide broader context.
• 7. Answer user queries by retrieving connected evidence and presenting it alongside the answer.

Evidence

KEYSTONE is an Evidence-First AI Company Brain built with Python, Streamlit, and FalkorDB.

Source

KEYSTONE flags these claims as a potential disagreement and encourages the user to review the underlying evidence.

Source

When asked about the launch, the prototype returns both conflicting dates and their source document instead of pretending that the evidence establishes one definitive answer.

Source