How To Guide: Semantic Question Answering

  • Create embeddings on your existing search index
  • Semantic search on OpenSearch as vector store
  • Pass the top results as context to ChatGPT (LLM) to generate AI answer from
  • Build SearchBox as the question answering UI
Semantic QA How-to hero

Interactive demo of semantic question answering based UI that we will build in the following steps

ReactiveSearch Features Used

Pipelines

Pipelines

Power the most demanding search workflows with JavaScript and a declarative authoring format

Learn More
OpenSearch

OpenSearch as vector store

OpenSearch as the search index and vector store, and the process of enriching an index with embeddings

Learn More
AI Answer

AI Answer

Allow users to ask natural language questions, provide a direct answer along with relevant search results

Learn More

Step 1: Starting with an existing search index, this step uses the vector indexing script to re-index the data with vector embeddings added. Works with an Elasticsearch or OpenSearch index

Indexing script is available over at appbaseio/ai-scripts.

Step 2: Author the search backend with ReactiveSearch pipeline

In this pipeline, we will show how to generate embedding for user's search query, search on OpenSearch as the vector store for finding the closest documents and generate an AI answer only from the result set.

Step 3: Build the search UI with ReactiveSearch UI kit

In this step, we will create a unique searchbox, define FAQs and connect the two. Next, we will take the SearchBox UI component's public demo as the base to modify and enable AI answering.

Try Semantic Question Answering

Get a 14-day free trial

Get Started