Ingestion
The document ingestion pipeline — upload, antivirus scan, parsing, chunking, labelling, embedding, and storage — with supported formats and limits.
Knowledge Base
Managing the GENIE.AI knowledge base — ingesting documents, the processing pipeline, the label taxonomy, document lifecycle, and content guidance.
GENIE.AI answers only from the documents you load into it. The knowledge base is the curated set of source documents that retrieval searches and the LLM reads from. This section is for the people who own and curate that content — knowledge managers, domain experts, and administrators — not developers of the pipeline internals.
If you want to understand how retrieval works under the hood, read the RAG Pipeline section first. This section is about the operational side: how documents get in, how they are processed, how to organise and maintain them, and what makes content retrieve well.
Pages in this section
- Ingestion — the upload → scan → chunk → label → embed → store pipeline, supported formats, and size limits.
- Document lifecycle — upload, processing states, updating, and retracting documents.
- Labelling & taxonomy — how chunks are labelled, the service-category taxonomy, and label-based filtering at retrieve time.
- Content guidance — what makes knowledge-base content retrieve well: document structure, chunking, and the effect of contextual retrieval.
How documents flow through the system
Upload (UI/API)
→ Document repository (ClamAV virus scan, storage)
→ Dataprep: docling parse → chunk → LLM label → context prefix → embed
→ ArangoDB (chunks + embeddings + graph nodes + labels)
→ Retrievable at query time
Every document is tracked by file_id, with per-chunk progress visible in the
ingestion log. A document can be retracted at any time, which removes its
chunks, embeddings, and graph nodes cleanly.
Who this is for
| Role | What you’ll do here |
|---|---|
| Knowledge manager | Upload and curate documents; organise the taxonomy. |
| Domain expert | Review and refine labels; judge retrieval quality. |
| Administrator | Monitor ingestion health; retract outdated content; size the knowledge base. |
Document Lifecycle
Upload, processing states, updating, and retracting documents in the GENIE.AI knowledge base.
Labelling & Taxonomy
How chunks are labelled against the service-category taxonomy, and how labels filter retrieval to keep answers on-topic.
Content Guidance
What makes knowledge-base content retrieve well — document structure, chunking, tables, and how contextual retrieval and labelling interact with content quality.