How-to guides¶
How-to guides are task-oriented. They answer "how do I do X?" and stop when X is done. They assume you already understand the concepts.
| Guide | Task |
|---|---|
| Write a compiler profile | Design and register a profile for your corpus |
| Tune extraction quality | Adjust volume, vocabulary, evidence grounding, SHACL validation, and all other quality levers |
| Use document distillation | Reduce token cost pre-fragmentation with six configurable distillation strategies |
| Control extraction focus | Tune precision vs recall with comprehensive, high_precision, and facts_only modes |
| Use batch extraction | Group fragments into a single LLM call to reduce API costs |
| Evaluate with Wikidata | Measure extraction quality against Wikidata ground truth; analyse recall gaps |
| Add a custom extractor | Ship a first-party extractor via entry points |
| Add a custom parser | Support a new document format |
| Connect a new source | Pull documents from an API or message queue |
| Configure Label Studio | Set up the human review loop |
| Run incremental recompile | Recompile only what changed |
| Explain a compiled artifact | Trace a fact back to its sources |
| Explain a conflict | Diagnose and resolve contradictions |
| Manage tenants | Full tenant lifecycle with GDPR erasure |
| Deploy on Kubernetes | Production Helm deployment |
| Rotate secrets | Update credentials without downtime |
| Bulk reprocess a profile | Re-extract all sources for a profile version |
| Generate an SBOM | Produce a software bill of materials |