Intelligent Q&A with RAG

Employees can ask any question about their company’s policies, projects, or knowledge base; the system understands context, retrieves the most relevant data (thanks to Elastic’s hybrid search)

Comment

Inspiration

Modern enterprises are overwhelmed with data across emails, wikis, documents, and chat logs. Searching for precise, context-aware answers from this vast knowledge base is essential but challenging. We wanted to reimagine how employees interact with their company’s information by combining Elastic’s hybrid search with Google Cloud’s generative AI, enabling natural language Q&A and intelligent, conversational experiences.

What it does

Intelligent Q&A with RAG is an enterprise-grade assistant that lets users ask any question about their organization’s data. It uses Elastic’s hybrid search to retrieve the most relevant documents, then leverages Google Vertex AI (Gemini) to generate concise, context-rich answers. The solution supports conversational follow-ups, cites sources, and adapts to the user's context, transforming both daily workflows and business processes.

How we built it

  • Data Ingestion: We loaded sample enterprise documents (wikis, PDFs, tickets) into Elastic, generating vector embeddings via Vertex AI Embeddings API.
  • Hybrid Search: When a user asks a question, the system gets both keyword and vector embeddings of the query, searching Elastic for top results using hybrid search.
  • Generative AI: Retrieved documents are sent to Gemini (Vertex AI) for answer synthesis, providing natural, context-aware responses with references.
  • Conversational UI: A chat-based frontend (using React/Streamlit) allows users to interact fluidly, ask follow-ups, and review cited sources.

Challenges we ran into

  • Integrating semantic and keyword search in Elastic for optimal relevance.
  • Ensuring generative answers are grounded in retrieved documents (minimizing hallucinations).
  • Managing embeddings lifecycle and updating the index efficiently.
  • Handling follow-up questions and maintaining context across conversations.

Accomplishments that we're proud of

  • Seamless integration between Elastic’s hybrid search and Google Vertex AI.
  • Real-time, context-rich Q&A over unstructured, enterprise-scale data.
  • Support for conversational context and source citation—boosting trust and usability.
  • Scalable architecture deployable on Google Cloud.

What we learned

  • Hybrid search significantly improves result relevance compared to keyword or semantic search alone.
  • RAG (Retrieval-Augmented Generation) frameworks can transform enterprise knowledge management.
  • Elastic and Vertex AI are highly complementary for building intelligent, interactive applications.

What's next for Intelligent Q&A with RAG

  • Add support for multimodal data (audio, images) and multilingual queries.
  • Enhance user authentication and data privacy controls.
  • Integrate feedback loops to continuously improve retrieval and generation quality.
  • Pilot deployment with real enterprise datasets and expand APIs for workflow automation.

Built With

Share this project:

Updates