VAST InsightEngine

VAST InsightEngine is the latest feature from VAST Data, developed to significantly simplify the construction of AI utilization infrastructure, which can often become complex. Traditionally, building data pipelines such as RAGs has required complex implementations across multiple servers. Using InsightEngine, implementation can be simplified on VAST Data, accelerating the use of AI with internal data.

Challenges in using AI with internal data

As AI technology rapidly evolves, there is a growing trend to reuse the vast amounts of data accumulated within companies and use AI to extract insights from them.As AI technology rapidly evolves, there is a growing trend to reuse the vast amounts of data accumulated within companies and use AI to extract insights from them.However, in reality, there are challenges to building systems such as RAG, including the following:

• Technical complexity

There are a wide range of component technologies involved, such as data preparation, vectorization, and messenger construction, so it takes time to launch.

• Rising cloud costs

For example, costs can quickly increase when vectorizing data in the cloud.

• Restricting document access scope

Because the documents that can be referenced differ depending on each employee's position and authority within the company, unrealistic measures such as providing a dedicated chatbot for each individual are necessary.

Especially in the construction of the RAG system,

1) Building a data pipeline (different processing systems are required for each type of data)

②Improvement of RAG accuracy (Embedding model selection, chunk design, etc.)

However, since step 1 requires a considerable amount of time and cost, it is difficult to focus on step 2.

VAST Data Insight Engine

VAST Data InsightEngine supports vector DB and messenger functions within VAST Data, making it possible to simply build a RAG system on a single VAST Data platform.

When building a RAG system on-premise, it is necessary to build, manage and link a vector database such as Milvus or Faiss, and a messaging system such as Apache Kafka using OSS, but this implementation is costly and time-consuming (see diagram below).

When a user transfers data to the VAST Data platform, VAST Data automatically and instantly communicates with the external embedding model, vectorizes the data, and stores it in the VAST Data vector database. As a result, when building a RAG system, all you need to do is:

・VAST Data Platform

・GPU for running LLM

This makes it possible to build a simpler system.

Adoption example

The VAST Data InsightEngine has been adopted by the NHL (National Hockey League), which stores all of the past 500,000 hours of archived data in VAST Data, and passes all of the data to Video Search and​ ​Summarization (VSS), an NVIDIA AI Blueprint, for analysis.

This makes it possible to instantly search and extract the scenes you want to see from the vast amount of archived data, greatly improving the fan experience and the quality of coaching.

https://www.vastdata.com/customers/nhl

Contact Us