MVP 1 Process mining + Knowledge base customer 1

[miguelgfierro]

Target Module

core

Problem Statement

Early version of Process mining + Knowledge base applied to one customer

Proposed Solution

Process mining

• Not structured data in sharepoint (pptx, docs, pdf, txt, md...) -> connection from sharepoint to knowledge graph
• Structured data in sharepoint (csv, xls, bbdd) -> connection a normalized db using the skill discover schema, maybe link it with a studio so it's more visual -> output goes to duckdb/sqlite (memory db not mantainence)
• Arquitectura hexagonal genérica, el primer adaptador es sharepoint.

Knowledge base

• LightRAG -> markitdown, configurable custom extraction. Ex: extract profiles of people in the data as a postprocessing of graph. Then it's optimized and indexed and the queries are very fast
• Queries for benchmark
• Tools for the agent to query the structured db (SQL query) and to the KG and and combine them. User through a tool that generates SQL code and that has RBAC permissions. Example: "Why the process in X region is not working as expected?"

Customer 1 needs

• I want to do a end to end analysis with a number of skills and/or agents -> mapping of the processes that are going to be executed. This is the analysis procedure.

No response

Additional Context

No response

[miguelgfierro]

Second phase studio local

Third phase K8S with deployment in studio, persistent db

[miguelgfierro]

When new files are added to shapoint, we need to identify it to trigger the extraction. -> TODO: Miguel

[miguelgfierro]

Adjacent proposal in #83: a generic data ingestion module with SharePoint as the first concrete adapter. Independent of the framing in this issue, but the SharePoint adapter and the hexagonal layout could serve as foundations if useful here.

[miguelgfierro]

working in https://github.com/fireflyframework/fireflyframework-agentic/releases/tag/v26.05.11