1. Defining the Action in UI Design

• Log in to the Kuika platform.
• Open the project you will be working on from the Apps screen.
• Go to the UI Design module and select the relevant screen.

2. Add Action

• Open the + ADD ACTION menu in the Properties panel on the right side.

• From the + ADD ACTION menu, add the Searching> Search action according to the trigger event you want (Initial Actions, OnClick, OnBlur, etc.).

3. Configure Action Parameters

• Search Type
  
  • Specifies the search type. One of the three options must be selected:
  • Vector Search: Vector Search converts the user's query and documents into multidimensional vectors and finds the closest content based on semantic similarity. With this method, even if the words used in the query do not appear in the document, semantically similar content can be identified.
    
    • Example: The query “how to improve customer satisfaction” may return a document titled “ways to improve user experience.”
  • Exact Vector Search: Exact Vector Search works similarly to Vector Search, but applies stricter threshold values to matches between the query and the document in order to increase accuracy. This ensures that only documents with very close (almost identical) semantic matches are returned as results.
    
    • Preferred for search scenarios that require greater precision and low error tolerance.
  • RAG Search (Retrieval-Augmented Generation): Finds the documents closest to the user's query using vector search and transfers the documents to the artificial intelligence model to generate a response in natural language.
    
    • Example: Query: “What is the old name of Izmir?”
    • Optional Parameter: Session ID
      
      • The user can add a Session ID parameter.
      • When Session ID is added: RAG Search results on Qdrant are associated with the Context information belonging to that session. Thus, questions asked in the same session are answered taking into account the context of previous messages.
      • When Session ID is not added: Message history is not kept. Each query sent is processed independently and only a response specific to that query is returned.
      • Both the platform and on-premise have been tested with artificial intelligence models (Qwen 2.5, Mistral, Deepseek, Llama).
      • The user can enter any text, number, or other type of value for the Session ID; there are no restrictions.
• Query
  
  • The textual expression to be searched for.
  • Example: “Where will the technology conference taking place on September 18 in Izmir be held?”
• Limit
  
  • The maximum number of results to be returned.
  • Applies only to Vector Search and Exact Vector Search.
  • If not specified, the default is 5 results.
• Collection Name (Optional)
  
  • The name of the specific collection to search.
  • If specified, the search is performed only in this collection.
  • If not specified, the search is performed in all collections.

Vector Search

• Purpose: Find documents containing the word “Izmir” or similar expressions
• Search Type: Vector Search
• Query: “izmir”
• Limit: 5
• Results are sorted from closest to furthest.

Exact Vector Search

• Purpose: Find documents that contain the word ‘İzMir’ exactly as it appears
• Search Type: Exact Vector Search
• Query: “İzMir”
• Limit: 5
• Case sensitivity must be taken into account.

RAG Search (Retrieval-Augmented Generation)

• Purpose: To obtain the most appropriate answer to a specific question, supported by documents
• Search Type: RAG Search
• Query: “What is the old name of İzmir?”
• This mode uses natural language processing (NLP) with OpenAI integration to generate responses.

• Dynamic Queries: The query value can be dynamically obtained from sources such as user input or form inputs.
• Collection-Based Search: Specific searches can be performed in separate collections rather than in the same data set.
• Using RAG Search, you can develop a question-answer assistant that works on specific documents or sources. This assistant answers the user's questions based on the given sources.

Technical Risks

• Incorrect Search Type: If the wrong Search Type is selected, appropriate results may not be returned.
• Missing Query Value: Empty queries may return errors or missing data.
• If RAG Search is to be used, a valid OpenAI configuration must be defined in the system.
• If an OpenAI-based model type is to be used, the settings for the relevant model (API Key, model name, etc.) must be specified in the configuration.

Search Action allows you to build advanced information search and recommendation systems on the Kuika platform. Thanks to the AI-powered RAG (Retrieval-Augmented Generation) feature, natural language responses can be generated to user queries based on relevant documents. In addition, Vector Search provides results based on semantic similarities, while methods such as Exact Search support classic, keyword-based information access. These classic methods are used to capture cases where specific words or phrases appear directly in documents. With all these capabilities, you can create smart and contextual search solutions that enrich the user experience.