Knowledge Management and Internal Data

The ingestion, structuring, and optimization of proprietary enterprise data to support accurate, fact-based AI responses.

Agents are only as effective as the knowledge they possess. The Knowledge Management (KM) section is the system you use to equip your Agent with proprietary, internal data—making it a knowledgeable expert on your organization's specific policies, products, and processes.

Effective KM involves three steps: ingestion, categorization, and optimization.

1. Uploading and Categorizing Internal Data

The Mattr platform allows for the bulk upload of internal documents, ensuring your Agent has a comprehensive knowledge base to draw from.

• Supported Document Types: Uptiq supports multiple document types (e.g., PDF, DOCX, TXT) and continuously expands this support, ensuring most internal documents can be ingested.

• Structured Categorization: Upload documents into specific, categorized buckets within the Knowledge hub (e.g., Lending, Compliance & Regulation, General Docs). This categorization is crucial for organizing vast amounts of data and improving retrieval accuracy.

• Knowledge Pipeline Optimization: Mattr uses specialized knowledge pipelines to index complex enterprise documents (e.g., those with graph or tree structures), increasing the accuracy of retrieval for technical data.

2. Optimizing Knowledge for Agent Performance

To meet the high-speed requirements of specific automations, you can designate data for rapid access:

• Flash Knowledge: For applications that demand instantaneous responses (e.g., conversational voice agents), designate certain documents as Flash Knowledge. These files are optimized for lightning-speed retrieval, making them essential for Agents running in a high-speed Flash Mode to prevent performance latency.

Flash Knowledge is intended for short-term reference and its retrieval speed is prioritized, meaning accuracy may depend more heavily on the source material quality.

3. Testing and Refining Knowledge Accuracy

For developers and advanced users, the platform provides tools to rigorously test how the Agent retrieves information from the ingested documents, ensuring the Agent is factual and reliable.

• Knowledge Hub Playground: This sandbox allows you to enter a query and immediately test the retrieval accuracy against your uploaded documents.

• Review Citations: The platform displays the Knowledge Chunks and Citations—the exact sections of the document used by the Agent to formulate its answer, along with a confidence score. This provides full traceability and validation of the response source.

• Advanced Refinement: If the Agent retrieves incorrect information, advanced users can directly intervene to improve accuracy:

• Edit Tags: Modify or correct the internal tags assigned to specific document sections.

• Split Sections: Break a combined section into smaller, more focused chunks to improve the context for retrieval.

• Re-index: Re-process the document after making changes to apply the new tagging and structure across the knowledge base.