TigerGraph Unveils CoPilot, Version 4.0, and New CEO

(AI generated by Shutterstock)

TigerGraph is holding its Virtual Graph + AI Summit 2024 later this week. But first it made a trio of announcements, including pair of pre-built CoPilots that turbocharge LLMs with graph data; the launch of TigerGraph Cloud 4.0, which separates storage and compute; as well as a new CEO, Hamid Azzawe, a Silicon Valley tech veteran who’s been brought in to grow the company.

TigerGraph is in the midst of a pivot in hopes of finding calmer water after some drama over the past two years. It’s moving away from being a pure-play graph database company, and moving towards being an AI-powered solutions provider that uses knowledge graphs. The industry focus won’t change dramatically (it intends to target financial services, where it already has a solid presence), but the makeup of the offerings will change.

The two new TigerGraph CoPilots unveiled today are part and parcel of its new identity as an AI-powered solution provider infused with graph tech. The first CoPilot, dubbed SupportAI, is geared toward enabling customers to have a better chatbot experience for customer support use cases, while the second new CoPilot offering, dubbed InquiryAI, is designed to let users gain natural language insights.

Graph Plus LLM

One of the downsides of using a generic pre-trained large language model (LLM) is that it doesn’t know anything about your company. Prompt engineering and retrieval-augmented generation (RAG) are techniques to “teach” the LLM what matters to you, but they don’t fully eliminate the tendency of LLMs to halluctinate, as tests have shown.

Graph databases speed the retrieval of connected data (ra2 studio/Shutterstock)

TigerGraph says the two new CoPilots will virtually eliminate hallucinations by centering input and output to and from an LLM on a customer’s own data. It does that by building a knowledge graph from the source material in the graph database and then applying standard RAG techniques to provide more contextual relevance to the LLM, Azzawe said.

“It’s far more relevant because not only do we do the better search on the components of the knowledge that is stored in the knowledge graph, but also when we bring in unstructured data, we’re able to extract concepts and concepts hierarchies that are domain-specific and enrich the knowledge graph with those concepts,” he explained in an interview with Datanami.

“As we assemble the context to feed into the LLM, now we’re able to do vector search as well as concept search. We find that before we send it to the LLM,” Azzawe continued. “Then the other part is once we get the response, which is really now far more accurate, we’re able to link it back to the knowledge base to allow the user to verify the response as well as explore the why and around it.”

SupportAI also looks for existing queries to help guide its interaction with the LLM. That “eliminates hallucinations completely,” Azzawe said.

TigerGraph is working on a CoPilot, dubbed QueryAI, which will allow business users to ask basic questions, such as “What accounts had the highest level of fraud?’ The CoPilot will then translate that natural language query into natural code that the graph database can execute (GQL, GSQL, or OpenCypher) and bring the answer back to the user. That CoPilot is still in development and is expected to debut in the second half of the year.

“With this, anybody could ask that question,” Azzawe said. “That’s the whole idea behind how democratizing insights from knowledge graph comes into play, for our CoPilots to be able to allow this to be available to anyone and everyone.”

Shrink-Wrapped Products

TigerGraph CoPilot architecture (Image courtesy TigerGraph)

The new CoPilots are shrink-wrapped products, or Solution Kits in the TigerGraph parlance, that can be used with customer’s existing TigerGraph implementations. Alternatively, they can be used as stand-alone products, as they have their own embedded knowledge base, Azzawe said.

In terms of the architecture, the CoPilots give customers flexibility to use them out-of-the-box or to integrate them with the vector database and LLM of their choosing. Out of the box, the CoPilots come with everything pre-integrated and ready to go. It leverages TigerGraph’s vector store, which functions as a disk-based “sidecar” that sits outside of the dataflow in the graph database cluster. It also uses the “house” LLM that TigerGraph uses, which is Meta‘s Llama 3. If users want to run their own vector database or LLM, they will have to do their own plumbing, Azzawe said.

TigerGraph Cloud 4.0 and New CEO

The company also unveiled TigerGraph Cloud 4.0, the latest release of the company’s parallel graph database that lives in the cloud as a managed service (all of TigerGraph’s offerings are also available as open source). The new database features full separation of compute and storage, enabling customers to scale the two elements up and down separately.

Azzawe was also introduced as the new CEO today. Azzawe is a database veteran who has had stints at IBM, Microsoft, AWS, Meta, and Bloomberg. He also founded his own startup, which used graph technology similar to TigerGraph.

“The reason I joined TigerGraph is because the technology was really appealing,” he said. “It was similar to the startup that I built from the ground up, but taking it to the next level in terms of graph databases.”

Hamid Azzawe is the CEO of TigerGraph

Azzawe, who was hired months ago as head of product and took the CEO position in February, takes the helm from Mingxi Wu, who managed the company from the middle of 2023. Wu, who returned to being head of engineering, was appointed interim CEO after the company ran into some financial issues related to its cash burn rate in 2022. The company’s founder and original CEO, Yu Xu, remains its lead engineer.

“Mingxi was interim CEO as part of a stabilizing the company and moving it to that next stage. And that’s again why he recruited me,” Azzawe said. “By the way, Mingxi and I go way back. He used to be an intern for me at Microsoft.”

Azzawe has a three-part focus for TigerGraph. The first is around “customer obsession” and focusing on making sure the product is mature, stable, and easy to use. “We’re now hyper focused on wanting to grow, and that’s where listening to customers and becoming customer-focus is paramount,” Azzawe said.

Number two is exploiting the company’s advantages at the intersection of generative AI and knowledge graphs, “having the only graph database that is built from the ground up to be enterprise-scale, hundreds of terabytes, and with real-time transactions.”

The third prong is offering AI-powered knowledge graph solutions that are more grounded in reality. “So yes, absolutely Mingxi started the transition towards enterprise maturity,” Azzawe said. “We’re taking it now to that next level.”

TigerGraph Virtual Graph + AI Summit 2024 is taking place May 1 and 2. You can sign up for the free event here.

Related Items:

New TigerGraph CEO Refocuses Efforts on Enterprise Customers

TigerGraph Cloud’s New Capabilities Help Close the Data and Decision Gap

TigerGraph Bolsters Database with Graph Analytics and ML