TigerGraph today made available in beta a copilot tool that takes advantage of graph technology and vector databases to make it simpler for end users to customize large language models (LLMs) using their own data.
Newly appointed Tigergraph CEO Hamid Azzawe said TigerGraph CoPilot is an AI assistant that leverages a knowledge graph developed by the company to make it simpler for end users to use natural language to expose data to an LLM via a vector database without having to master retrieval augmented generation (RAG) techniques.
Launched in advance of the company’s Graph + AI Summit, TigerGraph CoPilot builds a knowledge graph from source material and RAG techniques to improve contextual relevance and accuracy. It then determines the best available query needed to answer a question, runs the query, and returns the results in natural language, graph visualizations or in other data formats that may be required.
The overall goal is to streamline the process of sharing data with an any LLM without requiring the expertise of a data engineer, said Azzawe. That approach also mitigates hallucination issues by ensuring only curated queries are exposed to the LLM.
TigerGraph also plans to make the components it used to build TigerGraph Copilot available as open source software. It is optimized for TigerGraph Cloud 4.0, a database-as-a-service (DBaas) platform that runs on the Amazon Web Services (AWS) cloud. It is designed to scale by enabling IT teams to invoke compute and storage resources independently as needed. TigerGraph Cloud 4.0 also provides access to compute workspaces that share a common data store to enable IT teams to optimize the performance of different classes of workloads.
TigerGraph Cloud 4.0 can automatically detect formats, schemas and mappings, and provides connectors to data sources such as AWS S3 storage repositories as well as to data stored in Microsoft Azure, Google Cloud Platform, Snowflake repositories, or in open source formats such as Spark, Kafka and Postgres. The company is also making available solution kits for specific types of use cases within vertical industry segments to further reduce data management friction, said Azzawe.
It’s not clear the degree to which individuals will need to master prompt engineering as AI continues to evolve. In effect, Tigergraph is leveraging its knowledge graph technology to frame the questions an end user formulates in a way that can be used with any LLM. In fact, TigerGraph CoPilot can take advantage of change data capture tools to run a query any time data is updated, noted Azzawe.
That approach eliminates the need to rely on data engineers to formulate queries. Instead, data engineers can focus more of their time on managing backend services, said Azzawe.
Longer term, Tigergraph is working toward also enabling TigerGraph CoPilot to more proactively surface, for example, visualizations as data is updated, added Azzawe.
Ultimately, the ability to query generative AI platforms that have been exposed to additional data is about to be democratized. In effect, every organization will be able to instantly customize any LLM based on the data exposed to it without having to hire a small army of IT professionals to make it possible.
