Research topic

Local RAG and knowledge systems

Retrieval-Augmented Generation can make institutional knowledge accessible without surrendering control over documents, infrastructure or evaluation.

Documents, metadata, headings and chunks are prepared as an explicit information architecture rather than treated as raw model input.

Embeddings, vector stores and language models can run within organizational infrastructure when privacy and control matter.

Retrieval and generation are evaluated separately and jointly with realistic questions, missing-context cases and source checks.