Skip to main content

Knowledge Base | RAG

The Knowledge Base is a service based on RAG (Retrieval-Augmented Generation) offered on the Arvancloud AIaaS platform. The goal of this feature is to build a bridge between large language models (LLMs) and users' own data sources.

Standard language models are trained on public data and don't have access to an organization's internal, private, or very recent documents. The knowledge base feature lets AI models answer based on specific content controlled by the user.

How RAG Works​

This service is based on a multi-step process:

  1. Uploading documents: the user uploads their files (including text documents, PDF, Docx, and so on) to the system. These files are treated as the data source.
  2. Embedding: the documents' content is split into smaller pieces (chunking) and then converted into numeric vectors using an embedding model. Users can choose the embedding model they want based on the project's needs (such as accuracy or cost). These vectors are stored in a dedicated vector database (Vector Store).
  3. Retrieval: when the user sends a prompt or question, the system first converts it into a vector and searches the vector database to retrieve the most relevant pieces of information from the uploaded documents.
  4. Generation: the retrieved information is sent as context, along with the user's original prompt, to a large language model (LLM). The LLM then generates a response based directly on the information from the user's documents.

RAG Features​

The Arvancloud AIaaS knowledge base includes the following features:

  • Document management service: upload, manage, and delete the files that act as the knowledge source.
  • Embedding model selection: flexibility in choosing the model used to convert text into vectors, which affects the quality of semantic search.
  • Vector storage and management: automatic management of converting documents into vectors and storing them in a database optimized for fast search.
  • Retrieval service: the core of RAG, responsible for finding relevant information in the user's documents in response to a question.
  • Dedicated vector database (Vector Store): for each knowledge base, a dedicated database is created for you in the Arvancloud Database service panel. Besides creating, maintaining, and deleting this database automatically, you also get full access to it.

RAG Use Cases​

Using the RAG (Retrieval-Augmented Generation) mechanism, the knowledge base feature lets you connect large language models (LLMs) to your own controlled data sources. This enables many use cases where responses based on specific context and documents are critical.

Internal Knowledge Systems​

These use cases focus on empowering an organization's internal teams.

  • Q&A over organizational documents: employees can ask questions about internal policies, HR processes, technical guides, or financial documents, and get answers based on the company's official documents.
  • Technical assistant for developers: by uploading technical documentation, API guides, and internal code, engineering teams can ask complex technical questions and quickly find solutions or relevant code samples.
  • Training and onboarding: you can use this feature to create a training assistant for new employees. This assistant answers questions about onboarding processes, tools, and job knowledge based on training guides.

Customer Interaction Systems​

These use cases help improve the experience and support of end customers.

  • Customer support bots: create chatbots that, instead of generic answers, give precise and relevant answers to customers' questions based on product guides, the support knowledge base, and troubleshooting instructions.
  • Intelligent FAQ systems: turn a static FAQ section into a conversational interface that understands users' questions in different phrasings and extracts precise answers from product documents.

Information Analysis and Retrieval​

These use cases focus on processing and extracting information from large volumes of documents.

  • Legal document and contract analysis: lawyers and legal teams can use this system to quickly find specific clauses, obligations, or risks in long contracts or legal documents.
  • Research and development (R&D): researchers can upload scientific papers, patents, and internal test results, and use the system to summarize findings, compare methods, or extract key data.
  • Compliance review: organizations can upload documents on standards (such as ISO) or regulations (such as GDPR) to check whether internal procedures meet these requirements.