Skip to main content

Semantic & Vector Search in Adobe AEM with Viglet Turing ES

· 4 min read
Alexandre Oliveira
Alexandre Oliveira
Creator & Lead Developer, Viglet Turing ES

Once your Adobe Experience Manager content is indexed into an enterprise search engine, the obvious next step is to move beyond keyword matching. A visitor who searches "places to stay near the mountains" should find a page titled "Alpine lodges and cabins" — even though none of those words match. That's semantic search, and it's powered by embeddings.

This post builds on indexing AEM into Viglet Turing ES and shows how to add vector/semantic search and RAG on top — all open source, on your own infrastructure.

Keyword vs. semantic — what changes​

Classic search matches tokens. Semantic search converts both the query and your content into embeddings (high-dimensional vectors) and finds the nearest ones, so it matches on meaning:

In Turing ES this runs alongside the keyword index — you get faceted keyword search and semantic ranking on the same AEM content.

Step 1 — Configure an embedding model​

Turing ES talks to multiple embedding providers. In the admin console under Generative AI, configure an LLM instance and an embedding model:

ProviderEmbedding support
OpenAI✅
Azure OpenAI✅
Ollama (local models)✅
Anthropic, Google Gemini❌ (chat only — use one of the above for embeddings)

Using Ollama keeps embedding generation entirely on your own hardware — no content leaves your infrastructure, which matters for regulated AEM author content.

Step 2 — Pick an embedding store​

The vectors need somewhere to live. Turing ES supports pluggable embedding stores, including an embedded Lucene vector store that needs no extra infrastructure — ideal for getting started. For larger corpora you can point it at a dedicated vector database.

The embedded Lucene vector store is independent from the Lucene search-engine backend — you can run Solr for keyword search and still use the embedded Lucene store for vectors.

Step 3 — Embed your AEM content​

When AEM content flows in through the Dumont connector, Turing ES generates embeddings for each document and writes them to the store. Re-indexing an existing AEM site re-embeds it with your chosen model — no changes to the connector configuration.

Step 4 — Query semantically (and add RAG)​

With embeddings in place, Semantic Navigation ranks results by meaning, and you can layer RAG on top for grounded conversational answers over your AEM content:

# Semantic / RAG chat over the AEM index
curl "http://localhost:2700/api/sn/wknd-search/chat?q=Where+can+I+go+hiking+in+the+Alps"

Because retrieval is grounded in your indexed AEM pages, answers come back with citations — and, with Ollama, the entire pipeline (embeddings + LLM) can run on your own infrastructure.

Why this matters for AEM​

  • Better recall — visitors find the right page even when their words don't match the author's.
  • One platform — keyword facets, semantic ranking, and RAG over the same AEM content, not three separate systems.
  • Data residency — with local embedding + local LLM, AEM author content never leaves your perimeter.

Next steps​

Viglet Turing ES is open-source enterprise search with semantic navigation and generative AI. Index Adobe AEM, add embedding-based vector search and RAG, and keep your content on your own infrastructure.

ON THIS PAGE