Knowledge bases
Connectors let you find files by name. A knowledge base lets Rescale Edison read inside them, so you can ask a question rather than locate a document.
Browsing connected files and attaching them to a job work with the connectors included in Advanced Modeling & Simulation. Reading inside those documents needs Agentic Engineering. Your Rescale account team can confirm what your workspace has.
How this differs from browsing
A connector and a knowledge base point at the same storage but work at different levels. A connector works on the file system, matching names and paths, and hands back a list of files for you to open. A knowledge base works on what those files say, and hands back an answer.
Semantic matching is the part you notice first. A question about "thermal analysis results" reaches a document titled "heat transfer simulation outputs" even though the two phrases share no words, which is exactly the case where searching by name fails.
Asking a question
Open Edison and ask in plain language. For example: "What are the recommended boundary conditions for external aerodynamics?" Edison retrieves the relevant passages from your indexed documents and answers from them, rather than from general knowledge.
Every answer carries citations back to the source document and the location within it, so any claim can be traced to the file it came from. Opening a citation is worth the few seconds before you act on an answer, because it is the only way to tell an answer grounded in your documents from one that merely reads like it.
Supported sources and formats
A knowledge base can be built over an AWS S3, Azure Blob Storage or SharePoint connector. Databricks connectors do not support knowledge bases.
The document formats are the same whichever source holds them: PDF, including the images and tables inside it, Word, PowerPoint, plain text, and XML including PLMXML.
Setting up a knowledge base
An administrator does the first two steps once, and everyone in the workspace benefits from them.
- Connect the source, if it is not connected already. See setting up a connector.
- Click Sync to index its contents. Until a sync completes there is nothing for Edison to read.
- Ask Edison. Any member of the workspace can now ask questions against the indexed documents.
Scope what gets indexed
For SharePoint, admins can configure path restrictions to limit indexing to specific document libraries or folders.
Indexing an entire SharePoint site pulls in everything, including material that is outdated, superseded, or that you would rather Edison did not cite back to you.
Restricting the index to the libraries that hold current engineering documentation makes the answers better, because an answer can only be as good as the documents behind it.
Related
- Data connectors — the connection layer underneath
- Rescale Edison — where you ask the questions
- Metadata — the other way of making data findable