Dataset Reranking

ChatBotKit's dataset reranking option improves the accuracy of the records your AI retrieves. With reranking enabled, a dataset pulls more candidate records and applies a second-level sort, so the most relevant information is prioritized before it reaches the model. This works within retrieval-augmented generation (RAG), where the AI draws on dataset records while crafting its response.

The reranking algorithm acts as a second pass over the records that retrieval fetches first, scrutinizing them and ordering by relevance. Your AI still receives a limited number of records, but the relevance of those records is significantly higher.

What You Can Do

• Pull more candidates: Retrieve a wider set of records before narrowing down.
• Apply a second-level sort: Rerank the candidates so the most relevant ones come first.
• Improve answer quality: Give the model better context without increasing how many records it reads.
• Configure per dataset: Turn reranking on where retrieval accuracy matters most.

How It Works

Retrieval first fetches a set of candidate records for a query. With reranking enabled, ChatBotKit scores and reorders those candidates with a second-level algorithm, then passes the top, most relevant records to the model. Because the model receives the same limited number of records - just better ordered - quality improves without added cost to the response.

Setup

Open a dataset's configuration and enable the reranking option. Tune how many records are pulled before reranking to balance breadth against precision for your content.

Practical Uses

Reranking helps anywhere retrieval accuracy is critical: large knowledge bases, support bots that must cite the right policy, and research assistants where the best passage matters. By improving which records reach the model, reranking raises overall answer quality across your RAG workflows.