Embedding Nodes in AI+ Studio
Updated
Embedding nodes in AI+ Studio convert text into vector representations (embeddings) and use those vectors for semantic search and retrieval.
- An embedding is a numerical representation of text that captures its meaning and context.
- A vector index stores embeddings and enables similarity-based search across data.
Embedding nodes support:
- Semantic search
- Topic detection
- Knowledge retrieval
- Retrieval-augmented generation (RAG) workflows
How Embedding Nodes Work
A typical embedding workflow includes three stages:
- Generate embeddings – Convert input text into vectors
- Save embeddings – Store vectors in a vector index
- Retrieve embeddings – Perform similarity search to find relevant results
Together, these nodes enable you to:
- Transform text into embeddings
- Persist embeddings with metadata
- Retrieve semantically similar content
Generate Embeddings Node
The Generate Embeddings node converts input text into a vector embedding. Follow these steps to add Generate Embeddings Node to your deployment pipeline.
Click the ‘+’ button on the pipeline canvas.
Select Generate Embeddings under Sprinklr AI.
The node configuration screen will appear. Configure the fields as mentioned in the table below.
Configuration Fields
Field | Description |
Name | Specify a unique name to identify the node in the pipeline |
Input Variable | Select the variable containing the input text (for example, queries, topics, or documents) |
Provider | Select the AI provider used to generate embeddings |
Embedding Model | Choose the model used for embedding generation. This affects quality, dimensionality, latency, and cost |
Output Variable | Specify where the generated embedding will be stored |
Expected Result
- The system generates an embedding vector
- The vector is stored in the configured output variable
- The output is available for downstream nodes
Save Embeddings Node
The Save Embeddings node stores embeddings in a vector index, along with metadata. A vector index acts as a storage layer that enables efficient semantic search.
Key capabilities
- Select an existing vector index
- Create a new vector index directly from the configuration
This eliminates the need for separate backend setup.
Follow these steps to add Save Embeddings Node to your deployment pipeline.
Click the ‘+’ button on the pipeline canvas.
Select Save Embeddings under Sprinklr AI.
The node configuration screen will appear. Configure the fields as mentioned in the table below.
Configuration Fields
Field | Description |
Name | Specify a unique name for the node |
Embedding | Select the variable containing the embedding (typically from the Generate Embeddings node) |
Mapped Attributes | You can store additional metadata with each embedding using Mapped Attributes. Configure the following fields:
You can create multiple attributes as per your requirement. |
Vector Index | Select the index from the dropdown where embeddings will be stored. |
Inline Creation of Vector Index
You can create a new vector index directly from the Vector Index field during configuration.
In the Vector Index field, start typing the name of the index you want to create.
From the dropdown list, select Create.
The system creates the vector index immediately and uses it to store embeddings.
Expected Result
- Embeddings are stored in the selected vector index
- Metadata is saved alongside embeddings
- Data becomes available for semantic retrieval
Retrieve Embeddings Node
The Retrieve Embeddings node performs similarity search using embeddings.
- Compares an input embedding with stored embeddings
- Returns the most relevant matches based on similarity scores
Follow these steps to add Retrieve Embeddings Node to your deployment pipeline.
Click the ‘+’ button on the pipeline canvas.
Select Retrieve Embeddings under Sprinklr AI.
The node configuration screen will appear. Configure the fields as mentioned in the table below.
Configuration Fields
Field | Description |
Name | Specify a unique name for the node |
Select Embedding Variable | Choose the embedding used for similarity search |
Vector Index | Select the index to search for similar embeddings |
Confidence Threshold | Define the minimum similarity score required to return results |
Max Predictions | Specify the maximum number of results to return |
Retrieved Results Collection | Output variable to store retrieved results |
Expected Result
- The system retrieves the most similar embeddings
- Results are stored in the configured output variable
- Retrieved data can be used for downstream processing
End-to-End Workflow Example
Follow this sequence to build a semantic search pipeline:
- Use Generate Embeddings to convert input text into vectors
- Use Save Embeddings to store vectors and metadata in a vector index
- Use Retrieve Embeddings to find similar content based on input queries
Best Practices
Follow these best practices to improve accuracy and performance:
- Use meaningful, high-quality input text to generate better embeddings
- Store relevant metadata to enable filtering and traceability
- Reuse vector indexes across related workflows to improve efficiency
- Tune the confidence threshold to balance precision and recall
- Use consistent naming conventions for nodes and variables