AI Consultant VK Cloud
The AI assistant is a bot in MAX and Telegram based on artificial intelligence. It answers questions about the VK Cloud platform services, relying on the public documentation. The bot understands natural speech, and you can communicate with it as with a human.
The AI assistant works only with the public documentation and has no access to VK Cloud internal systems. For example, it cannot answer questions about the status of your requests to technical support.
-
Go to the AI assistant via direct links in MAX or in Telegram, or find it manually by entering
@vk_cloud_support_botorVK Cloud AI assistantin the search bar. -
Ask the bot a question as if you were talking to a human. The more detailed you describe what interests you, the more accurate the bot's answer will be.
To get the most useful and complete answer, follow these recommendations:
-
Ask questions one at a time. If you ask several questions in one message, the bot may fail to find answers to all of them in the documentation at once.
-
If possible, specify which VK Cloud service your question relates to. This gives the bot additional context and makes its answer more accurate.
-
The bot uses the terminology adopted in the VK Cloud public documentation. If possible, use the same terminology in your questions.
Example: answers to questions about Arenadata DB will be more relevant than answers to questions about Greenplum. The Arenadata DB service is indeed based on Greenplum, but in the documentation it is called Arenadata DB.
-
Avoid slang and abbreviations; the bot may not understand them. For example, the documentation for shared networks does not use the term "shared networks" (slang), and "legal entities" is not shortened to colloquial forms.
-
-
To make sure the answer is correct, check the documentation links attached to it. If all the links are irrelevant, the answer will not be complete and accurate enough. In this case, try to clarify the question: describe the problem in more detail and specify the subject area and terms.
-
Rate the usefulness of the answer by clicking one of the suggested buttons. Your feedback will help improve the AI assistant.
If the AI assistant could not help you, contact technical support using the contact information provided in the chat.
- Search for information in the VK Cloud documentation taking into account the context of the question. The AI assistant provides detailed answers and links to the source sections.
- Write Terraform scripts to solve various tasks, for example, to create a Kubernetes cluster.
- Advise on problems when using VK Cloud services. For example, the bot will help you deal with errors when connecting to virtual machines.
- Explain how to build and implement solutions based on the VK Cloud platform. For example, the bot can tell you how to build a Lakehouse solution or how to make an application fault-tolerant.
- Provide instructions on how to place applications in the VK Cloud store.
The AI assistant uses an LLM (Large Language Model) — a neural linguistic network trained on a large amount of data to understand and process texts in natural languages. To interact with the LLM, the AI assistant uses the RAG (Retrieval Augmented Generation) method, i.e., it supplements the user's question with context — a set of texts obtained by searching for relevant information in the VK Cloud public documentation.
Preliminarily, the VK Cloud documentation text is split into chunks (articles or their parts expressing a complete thought or a chain of thoughts). Each chunk is associated with an embedding — an ordered set of numbers calculated by a specific algorithm. For more information about embeddings, see the article.
Here is what happens next:
- The user asks the AI assistant a question.
- The embedding of the question is calculated, and the embeddings in the VK Cloud documentation that are closest to it "in meaning" are selected.
- The context is formed from the text of the chunks corresponding to the selected embeddings.
- The context is combined with the user's question and passed to the LLM input.
- The LLM returns an answer, which is displayed in the chat.
An LLM can solve a wider range of tasks than those of the AI assistant, for example, generate a marketing text or help a student with homework. To prevent misuse of the AI assistant, a multi-level protection system has been developed, which is based on an ML model (a machine learning model). A sign that the protection has been triggered is a message from the bot asking you to rephrase the question or description of the problem.
If you believe your question is correct and the protection was triggered falsely, report it to the bot administrator. This will help further train the ML model and reduce the number of false triggers.