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
Adapting search to your domain
Explores lightweight training of an enterprise search model using synthetic examples, testing whether adapting retrieval to a company’s vocabulary improves the passages found and the grounding of generated answers.
Read on this pageAt a glance
An enterprise AI answer depends on finding the right source material first. General search models may not recognise a company's specialised vocabulary or the way its employees ask questions. This paper studies whether a lightweight adaptation of the retrieval model can improve that fit in Corporate GPT.
The study trains a small set of model adjustments using generated questions and challenging alternative passages from the domain corpus. It evaluates the change both as a search-ranking problem and in the full question-answering pipeline, keeping the rest of the system fixed. The work gives enterprise teams a way to examine search quality separately from answer generation, and to test whether better source selection improves the grounding of the final response.
What the paper covers
Domain vocabulary can be addressed at the retrieval stage rather than only in answer prompts.
The study changes the embedding model while holding the surrounding pipeline fixed.
Offline ranking and end-to-end answer evaluation provide complementary evidence.
Research context
The reported gains come from the paper's domain corpus and evaluation setup, including automated judging and an offline held-out document. They are a case study rather than a guarantee for other customer corpora; each adaptation needs its own retrieval and answer-quality checks.
THE ORIGINAL WHITEPAPER
Download PDF Read it here.
The complete paper, with its original figures, tables and references.
12 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.