娇色导航

Keith Shaw: Hi everybody, welcome to DEMO, the show where companies come in and show us their latest products and platforms. Today, I'm joined by Cassie Shum. She is the Vice President of Field Engineering at RelationalAI. Welcome to the show, Cassie. Cassie Shum: Thank you for having me.

Keith: All right. Tell us a little bit about RelationalAI and what you’ll be showing us today. Cassie: Sure.

A little bit about RelationalAI — we are a relational knowledge graph, hence the word “relational” in our name. It’s a knowledge graph that is brought directly to the data cloud. In this case, we’ll be showcasing it in Snowflake.

I’ll walk through a few diagrams to show what the knowledge graph is and how it operates within the Snowflake ecosystem. nd then, of course, AI — where does the AI part come in? How do we interact with the knowledge graph?

Keith: Were you doing a lot of AI before generative AI came along, or is this something that grew out of generative AI technologies?

Cassie: I would say the generative AI movement has accelerated a lot of the things we’ve done — which I’m very excited to show in a bit. But before that, much of the AI we were doing centered on predictive and prescriptive analytics.

I have amazing teams who specialize in mathematical optimization, graph neural networks, and deep learning modeling. We’ve been doing this work for quite some time, well before the generative AI movement.

Keith: Within the enterprise, what role is this mainly designed for? When I think of “relational,” I think databases. And with AI, I think data scientists. Is this designed for them, or more for developers, security teams… who benefits most?

Cassie: Hopefully, in the future, it will be for massive organizations across the board. Right now, we target enterprise organizations — specifically, chief data officers, data engineers, and similar roles. But with the convergence of AI and applications, we’re seeing a movement from application-centric architectures toward data-centric architectures.

Eventually, I believe this will also target application developers.

Keith: Tell me about the problems these people are having, and how RelationalAI solves them. Cassie: What we’re seeing now, especially with the rise of generative AI, is the question: how do you actually utilize AI to its fullest capacity?

The foundational problems remain the same as in the BI and analytics era: data cleanliness, governance, and lineage. Big enterprises still often don’t know exactly what’s in their data. If you want to leverage generative AI or agentic systems, they can only work with what your data contains.

The knowledge graph gives people within your enterprise a clear understanding of your actual domain, and allows them to query across that domain and its associated data.

Keith: So if a company isn’t using RelationalAI, are they back in the Middle Ages? Would they be buried under stacks of paper? Or are they trying other methods and failing?

Cassie: We’ve seen many enterprises hand-roll their own semantic layer — essentially, what the knowledge graph provides — but the results are often bespoke and cobbled together. This prevents full utilization of the data sitting in the data cloud.

Our goal is to create an ecosystem where all your data lives in one place, like Snowflake, with the knowledge graph integrated directly into that environment. This way, you can access all your data using your domain knowledge and business logic — which is pretty exciting.

Keith: All right, enough talk from me — let’s get into the demo. Cassie: As we discussed earlier, RelationalAI is really about the convergence of AI and apps. The real question is: what kinds of questions can we answer with AI?

For example: * Can I connect my new Sony TV to my home Wi-Fi network? — This can be answered today using retrieval-augmented generation (RAG) and vector similarity searches on documents. * What were my sales last May of Sony TVs in Boston?

— This involves querying your own data, often using text-to-SQL on tabular data. * How many Sony TVs will I sell in Boston next May? — Now we’re predicting the future, which is much harder. * What should I do to sell more TVs in Boston next May?

or How many should I order to meet demand? — These questions require advanced reasoning. These are the kinds of questions that drive business success, but they can’t be answered with simple AI models alone. That’s why I’m setting the stage for how RelationalAI works within the Snowflake ecosystem.

Cassie: In many companies, this type of reasoning has been handled before — for example, with demand planning or supply chain planning — but these often sit inside traditional applications or microservices. The problem is they’re scattered across multiple systems, creating sprawl.

Our goal is to bring all this reasoning and data together into one ecosystem.

Cassie: For today’s demo, I’ll start inside Snowflake to show how we answer basic queries on top of a knowledge graph we’ve created. Then I’ll show how we can use ChatGPT or another LLM to query the knowledge graph and ontology directly.

Snowflake has a marketplace, and RelationalAI was one of the first native apps available there. Once installed, we run our engines inside Snowpark container services. From there, you can open example notebooks and start experimenting immediately.

Cassie: In this demo, I’m modeling a retail customer’s ontology and integrating their Snowflake data into the knowledge graph. The ontology defines entities like “seller” and “buyer” and the relationships between them, along with rules that reflect the company’s specific business logic.

No two enterprises are the same — each requires an ontology tailored to its domain. Often, these domains are siloed across multiple teams. With RelationalAI, we can unify them into a single, coherent ontology.

Keith: Do you typically go into a company and help them build their ontology, rather than expecting them to do it themselves? Cassie: Traditionally, yes.

Our field engineering team, which includes ontologists, will help model the business. But now, with generative AI tools, I think we can make this process much more self-service. Using tools like Norma for conceptual modeling, we can create detailed ontologies and then verbalize them into plain language.

These verbalizations can be fed into ChatGPT, which can then answer domain-specific questions, suggest refinements, or reconcile multiple ontologies.

Keith: So it’s answering based only on the ontology you gave it? Cassie: Correct. It’s not searching the web — it’s working strictly from the provided ontology. That means the results are private and domain-specific.

Cassie: Ontologies are never static — they evolve with the business. With generative AI, you can iterate on them quickly, continually refining the model to keep pace with changing needs.

Cassie: In this example, we’re running queries to see the number of items sold at past events. We can also calculate gross transaction value (GTV) per event and use predictive models to forecast future performance.

The advantage is that we’re bringing the reasoning to the data, inside Snowflake, rather than moving the data to external systems.

Keith: For people who want to try this out, is the Snowflake Marketplace the best place to start?

Cassie: Yes — you can request and download it directly from the marketplace, and someone from our team will reach out quickly.

For more information, visit our website at relational.ai, where we have documentation, getting-started guides, example notebooks, and sample data to help you explore the platform.

Keith: Cassie Shum from RelationalAI — thanks again for the demo. Cassie: Thank you so much. Keith: That’s going to do it for this week’s show. Be sure to like the video, subscribe to the channel, and leave your thoughts below. Join us every week for new episodes of DEMO.

I’m Keith Shaw — thanks for watching.