Documents#
If your organization manages knowledge in the form of internal documents—such as operation manuals, guidelines, FAQs, and reports—you can train the agent on any document type to use as knowledge.Regardless of format or file extension, simply uploading a document is enough for the AI to convert it into a usable data format for learning—no additional preprocessing required.The Knowledge page manages two data types: Documents and Feedback.
Registered documents and feedback are used as knowledge sources when the AI generates answers.
💡 Documents & Feedback
Documents are structured resources that organize official knowledge and information within the organization—such as policies, procedures, product information, and work guidelines. They serve as reliable sources for the AI system to reference. By leveraging these documents, AI agents can answer organization-specific questions accurately and consistently. Connecting internal documents to the AI system enables employees to find information quickly and improve operational efficiency.Feedback (Q&A Sets)#
Feedback is a core mechanism for continuously improving the AI agent’s performance. When the agent provides an inaccurate or incomplete answer, experts can store a model response alongside the original question to build question–answer (Q&A) sets. These datasets are used to help the AI provide more accurate answers to similar questions in the future. Through the feedback system, the AI evolves over time to better align with organizational needs and expectations, increasing user satisfaction and improving responses to work-related queries.💡 RAG
RAG (Retrieval-Augmented Generation) is a technique where the AI first retrieves relevant external information and then generates an answer based on that information. In other words, instead of relying solely on its internal knowledge, the AI “looks up” reference materials before it responds. This approach helps control hallucinations and improves reliability.Key Characteristics of RAG#
1.
Retrieval: Finds relevant information from documents, databases, and knowledge stores based on the query.
2.
Augmentation: Incorporates the retrieved information into the answer-generation process.
3.
Generation: Combines retrieved information with the AI’s base knowledge to produce accurate, up-to-date responses.
4.
Source Attribution: Can present the underlying sources used to generate the answer, increasing trust.
Benefits of RAG#
Access to the Latest Information: Utilizes information created after the AI’s original training.
Higher Accuracy: Provides more precise answers when specific facts or figures are required.
Domain Customization: Delivers specialized answers using internal enterprise documents or domain-specific resources.
Reduced Hallucinations: Lowers the chance of the AI producing incorrect information.
Everyday Examples of RAG#
Enterprise Knowledge Search: For questions like “What’s our company’s vacation policy?”, the AI searches internal documents to provide accurate answers.
Technical Support: Suggests concrete troubleshooting steps based on product manuals and technical docs.
Research Support: Searches recent academic papers to deliver answers aligned with current research trends.
Legal/Medical Guidance: References relevant laws or medical literature to provide trustworthy information.
A Simple Analogy#
RAG is similar to an open-book exam: instead of relying solely on memory, the AI looks up references to produce more accurate answers. It’s also like asking a friend a question and having them check a book or the internet before responding.RAG complements traditional AI, making it possible to build systems that are more accurate and reliable.Registering Documents#
Documents can be registered in two ways.1️⃣ Upload Files Directly#
Select a document from your PC and upload it.
Supported file extensions:PDF, PNG, JPG, JPEG, PPT, PPTX, DOCX, DOC, XLSX, XLS
If you require support for additional formats, please contact the Technical Team. Korean document files are also supported.
STORM Parse uses VisionLLM technology to transform complex document elements into AI-friendly structures.
It accurately recognizes visual and structural information that standard parsers often miss, enabling the AI agent to deliver more precise answers.
2️⃣ Upload via Google Drive Link#
Google Sheet, Google Drive
If you need additional link types for training, please contact the Technical Team. When registering a Google Drive link, be sure to check the sharing settings:Set access to “Anyone with the link” so the agent can read and learn from the document.
Editing Documents#
If a registered document needs updates, follow the method that corresponds to how it was registered.1️⃣ Documents Registered as Files#
Delete the existing file and re-upload the updated version.
2️⃣ Documents Registered via STORM Parse#
Go to the document’s Detail page.
Click Edit Conversion Result.
Review the original document on the left, make necessary edits, and save the changes.
3️⃣ Documents Registered via Link#
Navigate to the Detail page of the linked document.
Click Update to retrain on the latest version of the document.
Modified at 2025-10-20 05:44:19
