Learning from use
Proposes a governed learning loop that turns user corrections and usage signals into improvements across knowledge, search, instructions, memory and model training, with evaluation and human approval before release.
Read the paperRESEARCH / ENTERPRISE AI
Keeping structure in enterprise knowledge
How Corporate GPT prepares spreadsheets, slides, scanned pages and other complex files for search, preserving document structure and building summaries so important context survives the journey into an enterprise assistant.
Read on this pageAt a glance
A search model cannot recover a table header or a chart value that disappeared before indexing. This paper examines the document-preparation stage of enterprise retrieval, where spreadsheets, slides, scanned pages, transcripts and code need different treatment to preserve their meaning.
Corporate GPT brings these formats into a shared document representation, splits content along its structure and creates document-level summaries. The paper traces specific information-loss problems and the repairs tested against real inputs, from broken nested lists to oversized tables and missing scanned text. It is relevant to anyone assessing AI answers across a mixed document collection: source preparation can determine what the assistant is able to find before the question or the answer model enters the picture.
What the paper covers
Document structure carries meaning that fixed text windows can lose.
Different formats need targeted handling for tables, images, code and timing information.
Document summaries complement passage search when users need to find whole files.
Research context
The evidence is a structural evaluation on a multi-format regression corpus. The paper does not yet quantify the repairs' effect on answer quality. It also records remaining gaps in summarisation, personal-data coverage, conversion visibility and preservation of some document structures.
THE ORIGINAL WHITEPAPER
Download PDF Read it here.
The complete paper, with its original figures, tables and references.
22 pages · Read without leaving this pageFROM RESEARCH TO YOUR WORK
Discuss this research Bring your questions.
Discuss the approach, evidence and relevance to your team with the people building it.