Use CALIBER
Knowledge bases
The practical guide for ingesting knowledge, improving retrieval quality, and connecting grounded content to assistant or workflow behavior.
Use this page when the question is practical: ingest source material, verify retrieval quality, and connect grounded knowledge to assistant or workflow behavior.
At a glance
| Task | Start here | Deep reference |
|---|---|---|
| ingest documents | confirm storage and extraction path | Knowledge bases architecture |
| improve retrieval quality | inspect chunking, embeddings, and reranking | Knowledge bases architecture |
| use knowledge in assistant answers | connect retrieval to the consuming surface | Aria assistant |
| evaluate grounded answers | score citation quality and answer faithfulness | Evaluation and test sets |
1. What knowledge bases are for in CALIBER
Knowledge bases give CALIBER a governed retrieval surface: versioned corpora, ingestion, chunking, embeddings, optional graph extraction, and retrieval that can be measured rather than assumed.
2. Common tasks
| You want to... | Read this next |
|---|---|
| upload and ingest content | Knowledge bases architecture |
| improve search or answer quality | Evaluation and test sets |
| debug a missing source or file path | Storage and state |
| reason about grounded-answer trust | Trust and governance |
3. Common failure modes
| Symptom | First thing to check |
|---|---|
| Files uploaded, but answers are weak | chunking, embeddings, reranking, or evaluation coverage |
| Sources exist, but retrieval cannot find them | ingestion status, storage path, or project scope |
| Answers cite the wrong material | retrieval quality, prompt instructions, or judge coverage |