It searches only where it should
It indexes only the content you choose, with the permissions of the person searching.
Semantic and agentic search that answers questions
Anchored to the official contentQuestions in natural languageOpen source
People type a question and find the passage of the page that holds the answer, even when their words differ from the site’s.
People no longer search by keyword.
Semantic and agentic search is a search engine for websites and portals that understands the meaning of a question and answers with the official content, citing the page. SparkFabrik is an AI Native Engineering company that designs and runs the conversational search of the Luiss website.
Users are used to asking an AI and getting an immediate answer. On a site with thousands of pages, classic search returns a list of results, often wrong when the words do not match.
People who cannot find what they need call, write or leave. And a generic assistant, plugged into the site with no rules, risks answering with what is not there.
A conversational search that answers on the page, precise because it is anchored to your content.
Finds by meaning
The right question,even with the wrong words.
Search understands what the question means: “duplicate card” finds the page about a new badge. The result is the passage that answers, with a link to the page.
Site searchInteractive example · sample pages
How do I request a duplicate card?
By keyword3 results
Library card: rules…the card is personal and…
Fee refund requests…how do I request a refund…
Student forms (PDF)List of forms…
By meaningthe answer
If it is lost or stolen, you can request a new badge from the student portal, in the Services section. You collect the new badge at the registry office.
Lost university badge ↗duplicate ≈ new badgecard ≈ badge
Try a question
Words and meaning
Results that saywhy they are there.
Hybrid search: lexical and semantic, with filters and sorting. Every result explains why it is there, across product sheets, documentation and editorial content.
Technical catalogueInteractive example · sample catalogue
valve for a solar system that withstands high temperatures
3 results
Thermostatic valve for solar circuits
Why it is here: “solar” in the title, and the sheet gives the maximum operating temperature, which is what you ask.
word: solarmeaning: high temperatures
Guide to choosing solar thermal components
Why it is here: it covers the heat resistance of components, even though it never says “valve”.
meaning
Solar thermal in summer: how to protect the system
Why it is here: it explains what happens to components when the circuit exceeds its design temperatures.
word: systemmeaning: high temperatures
You see what is missing
Unanswered questionstell you what to write.
A dashboard shows the people who manage the content the questions, the results and the unanswered questions, and shows which content is missing or badly written.
Editors' dashboardInteractive example · sample data
Unanswered questions, most frequent first
When do master's enrolments open?Missing content
Can I pay the second instalment in parts?Missing content
Where do I find the exam calendar?Unclear page
The situations people start from.
Where it helps most: universities, public administrations, technical catalogs, large editorial archives.
01 / 04
We have thousands of pages and nobody finds anything
Search understands the meaning of the question and finds the right content even when the words used differ from the ones on the page.
02 / 04
Our users type questions, not keywords
How do I request a duplicate card? Search answers with the passage of the page that holds the answer, and the link to the page.
03 / 04
The product catalog is huge and technical
Search by meaning across data sheets, documentation and editorial content, with filters and results that explain why they are there.
04 / 04
We want an assistant on the site, but a safe one
Agents and chatbots use search as a tool: they answer only with the indexed content, with guardrails on.
How it works.
A pipeline indexes the content, the search combines words and meaning, and agents know when to use it.
- Content
Indexing
CMS pages and documents turned into structured text and embeddings, updated when they change.
- Search
Words and meaning
A hybrid search on Typesense: lexical and semantic, with filters and sorting.
- Agents
When to search
An agent decides whether and how to use search. Other tools connect through the MCP client we maintain, enabled one by one.
- Control
Guardrails and perimeter
Answers only from the indexed content, moderated conversation, unanswered questions logged.
- Drupal
- Search API Typesense
- Typesense
- LiteLLM
- Cohere
- MCP
- AI Guardrails
- Kubernetes
Perimeter, guardrails, metrics.
A search with AI inside is governed like a production system.
Answers under control
The conversation is moderated and the answers stay inside the indexed content.
Visible quality
The dashboard shows which questions go unanswered.
Success Stories
In production.
The numbers come from the Luiss platform, built with open source modules we maintain.
Luiss
Conversational search over the university’s content, with Search API Typesense and RAG.University · In production
Modules
Search API Typesense and the Typesense RAG recipe for Drupal are maintained by us on drupal.org.Open source
Talk
AI and Typesense: how to integrate semantic search in Drupal, at Talks On My Machine.Event · ReplayWhat is always included.
The indexing
Your content, ready to be searched
CMS pages and documents, updated automatically when they change.
The search
Words and meaning
Configuration of hybrid search, filters, sorting and languages.
The interface
Where people search
The search component on the site or the portal, accessible and consistent with the design system.
The metrics
What works and what does not
The question log and the dashboard for the editorial team.
Three steps, in this order.
- 01
Discovery
We look at the content, the most frequent questions and the perimeter of the search.
You leave withThe list of questions the search must answer and of the content to index.
- 02
Build
Indexing, search and interface, tested on real questions.
You leave withThe new search live, with metrics in place.
- 03
Evolution
Agents, a conversational assistant, new sources and new languages.
You leave withA search that grows with the content.
See it on your own pages.
Bring us the questions you get most often. We show you what the search would find, and where from.
The services inside this solution.
A solution is made of our services, put together for one precise problem.
AI Development
Semantic search, RAG, agents.MCP servers
The agents that search, connected to other tools with permissions and tracing.Drupal
Search API and the modules we maintain.Strategy & Design
The search experience and the content architecture.Managed Services
Running it after the release.
AI Native Engineering
No lock-in to a vendor or a model.
Search runs on open source modules we maintain on drupal.org, such as Search API Typesense: no license and no lock-in to a vendor.
Frequently asked questions.
The questions you are already asking.
How is it different from the search I already have?
Classic search looks for words. This one looks for meaning: it finds the right content even when people use words different from the ones on the page, and answers with the passage that holds the answer.
Do you use AI on our projects? With what guarantees?
Yes. We build the search with the same method as all our projects: specification before code, tests and review, and a person who approves every release. It runs on open source modules we maintain, such as Search API Typesense, and indexes only the content you choose, with guardrails active on the answers.
Does it only work with Drupal?
Drupal is the most frequent case, with the modules we maintain. Indexing can also include documents and other sources.
Can it make answers up?
Search returns existing content. When there is a conversational layer, it answers only with the indexed content and with guardrails on.
Which languages does it support?
The ones on your site. Semantic search also works across different languages.
What does it cost?
It depends on the amount of content, sources and languages. The Discovery gives you the order of magnitude.
Talk to the CEO or the CTO.
Tell us what you publish and what you are asked most often. We will tell you where we would start.
Tell us about your project
Updated
