AI integration / Assistants and agents / ARTEFACT
Retrieval over your documents
A working retrieval layer over your own material, with the ingestion, chunking and permission rules written down.
Retrieval over your documents drawn as a pipeline. Parts: Ingestion pipeline, Search index, Answer path and Test report. Acceptance: The pipeline ingests every source in the approved register and reports any rejected file. A test user cannot retrieve material outside that user's recorded permissions. Every accepted retrieval case returns the expected passage with its source citation.
01
Ingestion pipeline
02
Search index
03
Answer path
04
Test report
- The pipeline ingests every source in the approved register and reports any rejected file.
- A test user cannot retrieve material outside that user's recorded permissions.
- Every accepted retrieval case returns the expected passage with its source citation.
WHAT THIS ARTEFACT IS
A named output with defined parts
Retrieval over your documents is a pipeline that ingests approved material, divides it into searchable units, applies access rules and supplies cited source passages to a model at answer time.
What is inside it
- 01
Ingestion pipeline
Approved files enter through a defined source, receive a stable identifier and pass format checks before indexing.
- 02
Search index
Text units, source metadata and access rules are stored together so retrieval can preserve document ownership.
- 03
Answer path
A query retrieves permitted passages, passes them to the model and returns citations to the source material.
- 04
Test report
Representative questions record the passages retrieved, the expected source and the observed result.
WHAT IT TAKES
Inputs, time and ownership
FROM YOU
4 inputs
- Access to representative documents
- Document ownership records
- A named content decision maker
- Existing retention and access rules
WORKING RANGE
CLIENT ROLE
A content owner must approve the source register, access rules and expected answers used in retrieval tests.
WHERE IT SITS