Knowledge Base¶
The knowledge base is DSAgt's catalog of domain knowledge — reference corpora and your own documents — that the agent searches to ground its work on scientific data-processing and AI-readiness evaluation.
Domain-knowledge collections¶
| Collection | Source | Populated by |
|---|---|---|
| Reference corpora | NeMo Curator + AIDRIN (data-curation and AI-data-readiness references) | dsagt init (chosen collections) |
| Your documents | Papers, standards, protocols, schemas you ingest | Agent's kb_ingest |
The has access to three knowledge base tools: kb_ingest (index a file or directory into a named collection — long ingests run in the background), kb_search (retrieve across one or more collections), and kb_list_collections (see what's indexed).
Hybrid vector search¶
Retrieval is hybrid — dense semantic embeddings fused with sparse BM25 keyword matching — by default per collection for focused retrieval:
- Semantic embeddings catch paraphrase and synonymy: a query about "missing values" finds a passage on "null rates" even with no shared words.
- BM25 keyword matching catches the exact terms embeddings tend to under-rank — identifiers, gene names, parameter flags, standard names — where a literal match matters.
- Per-collection partitioning scopes a search to a domain, so a materials-science query isn't diluted by genomics references.
- Optional cross-encoder reranking re-scores the top candidates for precision when it's worth the extra pass.
The default embedder is a local sentence-transformers model (~130 MB).
Shared vector store¶
The same vector store additionally supports DSAgt's memory (explicit + episodic), skills discovery (the installable-skill corpus), and code execution tracking (the code_use records). Each is a separate partitioned collection in that store, sharing one embedder and one ChromaDB. Respective docs on memory, skills, and provenance share specific details about those collections.
Setup¶
dsagt init sets up the knowledge base from your choices in the interactive menu.
Try it¶
Then, in the agent — substituting <your-docs-folder> with any folder of your
own documents (papers, protocols, schemas):
-
Ingest the docs in
<your-docs-folder>into a collection nameddomain. -
Search the
domainandaidrincollections for how to assess data completeness.