Fin-CX-Retrieval Model
A specialized AI model that scans knowledge sources to identify and retrieve relevant content. Uses semantic understanding rather than keyword matching to understand customer intent and select the top candidate content pieces for response generation.
Semantic Search vs Keyword Matching
Traditional search systems rely on keyword matching—finding documents that contain the exact words in a query. The fin-cx-retrieval model uses semantic understanding instead, grasping the meaning and intent behind questions even when phrased in different ways.
For example, a customer asking 'How do I get my money back?' and another asking 'What's your refund process?' are semantically similar, even though they share few keywords. The fin-cx-retrieval model understands this similarity and retrieves the same relevant refund documentation for both queries.
The Retrieval Process
When an AI Agent receives a customer query, the fin-cx-retrieval model executes a multi-step process:
- Understand intent: Analyzes the customer's question to grasp what they're truly asking, beyond surface-level keywords
- Search across sources: Scans all available content including help articles, snippets, macros, and knowledge base documentation
- Match on semantics: Identifies content that addresses the customer's intent, even if exact keywords don't match
- Select top candidates: Returns the most likely relevant pieces for further processing by the reranker model
Integration with RAG
The fin-cx-retrieval model is a critical component of the retrieval-augmented generation (RAG) pipeline. By pulling relevant content from knowledge bases, it ensures that AI Agent responses are grounded in actual documentation rather than hallucinated. This retrieval step happens in milliseconds, enabling real-time responses that maintain both speed and accuracy at scale.