Overview
A RAG (Retrieval-Augmented Generation) application in Dataworkz is a production-ready Q&A interface that answers questions using your organization's own data. When a user asks a question, the app retrieves the most relevant content from your connected data sources and passes it to a configured LLM to generate an accurate, source-backed response.
Unlike querying an LLM directly, a Dataworkz RAG application:
Grounds answers in your data — responses cite specific documents or records, not general model training
Shows source attribution — users can see exactly which document or chunk each answer came from
Supports metadata filters — queries can be scoped to specific document types, departments, regions, or any indexed attribute
Tracks response quality — per-query metrics (groundedness, context relevance, answer relevance) are available in the Insights panel
Supports A/B testing — the Experiments feature lets you compare retrieval configurations to find the optimal setup
The RAG Pipeline in Dataworkz
Every RAG application follows this pipeline:
Connect — link data sources via Configuration → Integrations (cloud storage, SaaS platforms, databases, streaming systems)
Ingest — bring data into the Data Store in Data Studio; Dataworkz extracts and organizes content into collections
Chunk & Embed — Dataworkz splits content into segments and converts each segment into a vector embedding using your chosen embedding model
Configure retrieval — define how the system selects relevant content at query time: top-N chunks, metadata filters, query rewriting, guardrails, and caching
Query — users ask questions via the Q&A interface; the app retrieves relevant chunks, assembles context, and generates a response
Two Creation Paths
Default settings — Dataworkz applies pre-configured chunking, embedding, and retrieval settings automatically. Best for getting started quickly or evaluating the platform.
Custom settings — a four-step wizard (LLM → Ingest → Chunk & Vectorize → Retrieval) gives you full control over each stage. Use this path when you need specific embedding models, retrieval strategies, filters, or guardrails.
Both paths are covered in the Quickstart guides.
Last updated