Most teams working with Elasticsearch, OpenSearch or RAG pipelines focus on ranking, embeddings or model quality when trying to improve relevance. But in many cases, the issue starts much earlier: in how text is normalized before indexing. In a previous post, we...
Some RAG issues have a simpler fix than people think: better text normalization. One common culprit is stemming. Stemming is a blunt, error-prone approach: it strips word endings mechanically, without properly accounting for morphology, part of speech, or context....
Recent Comments