Next Steps

Placeholder. Semantic-search-specific best-practices guides will be added over time. The cross-links below point at the right general guidance for now.

You’ve created a Search function, loaded a reference corpus, and queried it. From here, the work is mostly about curating the corpus, integrating queries into your application, and using similarity scores well.

Curate the corpus

Semantic search improves by improving the corpus, not by reviewing individual queries. The highest-leverage actions:

• Add coverage for queries your function currently answers poorly.
• Retire stale items that no longer apply.
• Re-segment items if they’re too large (results lack focus) or too small (results miss context).
• Deduplicate near-identical items that crowd out other relevant results.

Plan for production

• Pagination. Semantic search returns ranked lists. Decide how many results to show users and where the cutoff is.
• Score thresholds. Many applications hide results below a certain similarity score rather than showing low-quality matches at the bottom.
• Caching. Identical queries return identical results until the corpus changes. Cache aggressively if you expect repeated queries.

Related: Production Integration Patterns.

Go deeper on the platform

• Developer Platform — architecture, principles, and reference for building production integrations.
• API Reference — complete REST API documentation.

Or try another function type

• Create Your Own Classification Function — assign each input to one of a set of labels.
• Create Your Own Box Detect Function — find and locate a specific object in images.