Skip to main content

Retrieval Augmented Generation (RAG) agent

· 4 min read

RAG agents enhance Large Language Models (LLMs) by providing them with access to an external knowledge base, enabling accurate and efficient information retrieval from unstructured data.

RAG-agent1

Key features​

  • Knowledge base integration:

    • RAG agents leverage uploaded unstructured data (documents) within a workspace as a knowledge base.
    • Supported file formats: PDF, DOC, TXT, PPT.

  • Accurate information retrieval:

    • By grounding LLMs with a specific knowledge base, RAG agents provide more accurate and contextually relevant responses.

  • Efficient information retrieval:

    • Streamlined process for retrieving information directly from uploaded documents.

  • Data extraction and formatting:

    • Ability to extract specific details from documents in JSON or tabular formats.
    • Capability to compare information across multiple documents.

  • Document and Chunk referencing:

    • Provides references to the specific documents and chunks of information used to generate responses.
    • Current limitations: Preview is only available for PDF documents.

  • Custom RAG Agents:

    • Specialized agents for specific use cases, such as invoice analysis (Invoice Analyzer Agent) and clinical report analysis (Clinical Report Analyzer Agent).
    • These custom agents are designed to handle variations in document structures and extract relevant information.

Specific RAG agent capabilities​

  • Simple Question Answering:

    • Provides answers to direct questions based on document content.

  • Detailed Information Extraction:

    • Extracts specific details from documents, such as metrics, names, or values.

  • Comparative Analysis:

    • Compares information across multiple documents and provides a summary or conclusion.

  • Data Formatting:

    • Formats extracted data into JSON or tables.

Custom RAG agents​

  • Invoice analyzer agent:

    • Extracts data from various invoice formats, including tracking IDs and other relevant details.
    • Customized to generate outputs in data in JSON format and Table formats

  • Clinical report analyzer agent:

    • Extracts data from clinical reports, including test results and patient information.
    • Customized to generate response in JSON and table formats.

Workflow​

  1. Upload unstructured documents to the workspace.

  2. Select the RAG agent within the platform.

  3. Input your questions or requests related to the uploaded documents. The processing is as follows:

    • The system processes the documents (chunking, indexing). Processing time depends on the number and size of documents.

    • While the agent is processing, a message is displayed indicating the processing status.

  4. The RAG agent generates a response based on the information retrieved from the documents. The following references are made:

    • The system provides references to the documents and specific chunks used to generate the response.

    • The documents tab shows the exact chunk that the answer came from.

  5. The extracted information can be presented in JSON or tabular formats as requested.

Limitations and Future Enhancements:​

  • Document Preview:

    • Currently, document preview is only available for PDF files. Support for other file types is planned.

  • Direct chunk highlighting:

    • Future enhancements will include highlighting the specific chunk within the document when references are clicked.

  • Knowledge source integration:

    • Future development will explore the ability to connect to external knowledge bases and repositories.

  • Audio and video file support:

    • MP3 and MP4 support are on the roadmap.

  • API/SDK availability:

    • APIs and SDKs for RAG agents are planned for future release.

  • Custom agent building tools:

    • Tools to allow users to build and customize their own agents are planned.

  • Orchestrator copilot:

    • The ability to have the orchestrator copilot query both structured and unstructured data is a planned feature.

FAQ​

  • Can I upload documents in languages other than English?

    • Yes, the RAG agent supports multiple languages, leveraging the capabilities of the underlying LLM.

  • Does the default Data Conversion (DC) agent work with documents?

    • No, the DC agent does not work with unstructured documents. You must select the RAG agent for document processing.

  • Is there a limit on the number of documents I can upload?

    • There is no hard limit, but processing time will increase with the number of documents.

  • Can I switch between different agents?

    • Yes, you can switch between RAG agents and other available agents at any time.

  • Can I Delete a document?

    • Yes, documents can be deleted by selecting "View All" and then selecting the document to be deleted.