A schematic on a dark ground. Top row: the SharePoint wiki of 1,027 pages syncs through the Microsoft Graph API, using an Azure AD app registration, into a privacy filter and an OpenSearch index of 2,135 passages from 715 pages. Bottom row: a researcher asks the lab AI chatbot, which calls the MCP tool search_group_wiki, which runs hybrid semantic and keyword search fused by RRF over the index; the answer returns citing the wiki pages. A dashed outline marks the filter, index, tool and search as hosted at the university.

Agentic RAG · 2026–present

Wiki RAG: the lab's know-how on call

The lab's internal wiki answering questions in its AI chatbot, with the right page in the top 10 for 99 of 100 test questions.

The short version

A research group's working knowledge, which tool to use, which recipe works, how to fix a failed process step, sits in an internal SharePoint wiki of more than a thousand pages that few people search well. A generic AI chatbot cannot see it and would invent the answer. The lab needed answers drawn from its own pages, citing them, without the data leaving the university.

Role
Initiator and builder
Period
2026–present
PythonMicrosoft Graph APIAzure AD app registrationSharePointOpenSearchHybrid searchBM25Reciprocal Rank FusionMCPLLM evaluation

Schematic of the wiki RAG pipeline. Top row, left to right: SharePoint wiki, 1,027 pages; Microsoft Graph API with an Azure AD app registration, highlighted; a privacy filter that leaves restricted pages out; an OpenSearch index of 2,135 passages from 715 pages. An arrow runs down from the index to the bottom row, which reads right to left: hybrid search, semantic plus keyword fused by RRF; the MCP tool search_group_wiki; the lab AI chatbot on university-hosted models; and the researcher who asked, who receives an answer citing the wiki pages. A dashed outline marks the filter, index, tool and search as hosted at the university. The wiki reaches the index through the Microsoft Graph API; questions reach it through the chatbot's MCP tool, and nothing leaves the university.

How

  • Syncs the wiki through the Microsoft Graph API with an Azure AD application registration, the access route agreed with university IT; 1,027 pages crawled, privacy-restricted pages excluded.
  • Indexes 2,135 passages in OpenSearch and searches them two ways at once, semantic (kNN) and keyword (BM25), merged by Reciprocal Rank Fusion.
  • Served to the lab chatbot as an MCP tool on university-hosted models, so answers cite their pages and nothing leaves the university.
  • Measured, not assumed: 100 test questions written by an LLM from randomly sampled wiki passages, verbatim copies rejected, scored at passage and page level for each search mode.

Part of the lab's AI work alongside the agentic literature RAG, which the same chatbot uses for published papers.