Diagrid announces Dapr graduation and plans for LLMs

Bart: We should have use Pod Topology Spread Constraints to solve this.

Salaboy: What's your role? And where do you work?

Bart: Hello, my name is Mauricio Salatino, also known as Salaboy. I work for Diagrid, a company that is working on the Dapr project, which graduated this week and is expected to be announced in the keynotes.

Salaboy: What would you like to share today?

Bart: I would like to share with people what Dapr is about and why you should care. In the context of platform engineering, people tend to put a lot of tools together to help teams consume infrastructure and provision infrastructure. However, sometimes people forget about developers. Dapr brings a solution to help development teams build applications faster and make them agnostic of the infrastructure they require. Dapr does this by creating a sidecar that exposes APIs for applications to consume. Some of the APIs this sidecar exposes are for PubSub, Workflows, and Service-to-Service Communication, all of which implement resiliency, observability, security, and other features that companies care about. Developers often leave these features out or add them as dependencies, complicating their applications. After learning about Kubernetes and the basics of observability and security, I think enabling developers should be a priority in platform engineering.

Salaboy: What problem, what's the technical issue that Dapr is solving?

Bart: The technical issue usually involves how developers consume infrastructure. To consume infrastructure, you probably need an SDK, a client, or a library to connect to your databases, message brokers, and everything that is distributed, such as authentication and authorization services. Dapr solves the problem of connecting to these things by exposing APIs. This allows developers who know how to consume APIs to simply call APIs instead of adding dependencies to their code, making their application agnostic of the environment where it is running.

Salaboy: And can you share a bit about the before and after of this announcement?

Bart: Dapr is a very stable project that was donated to the CNCF in 2019. Since then, the Dapr project has been working on the maturity process. We were incubating and gained a lot of adoption from different companies. We are working on stabilizing all the APIs and components that implement these APIs to run them across multiple cloud providers and on-prem. Since the announcement happened this week, I've seen a lot of interested people asking what this is about. I see this project now facing a phase of explaining to people what the project can help with, how it relates to platform engineering, and why teams should look into it after they manage the infrastructure side of things.

Salaboy: Is Dapr open source and part of the CNCF landscape?

Bart: Dapr is an open source project, it's part of the CNCF, it's graduated now. This means it has gone through a maturity process that helps to validate there is an interest in the project, and there are some companies like Diagrid providing services around it.

Salaboy: And what's Diagrid's business model?

Bart: Diagrid offers tools for working with Dapr, not only for Kubernetes users. We offer monitoring tools, including a tool called Diagrid Conductor, which is free to use. If you're using Dapr, you can connect Diagrid Conductor to see what your applications are doing, how they interact with the infrastructure, and understand how efficient your applications are, as well as the structure and communication patterns they use. We also have a SaaS offering called Diagrid Catalyst, which offers Dapr as a service. This allows you to connect applications running anywhere, from on-premises to cloud services like Cloud Run or Amazon application services, and interact with the Dapr APIs, even if you're not running on Kubernetes.

Salaboy: Who are your main competitors?

Bart: We don't have direct competitors in the CNCF space. However, because we expose APIs, we are competing with specific infrastructure-related tools like Apache Kafka or Apache Pulsar, as well as tools in the workflow engine space, such as Temporal, since we also offer workflow capabilities.

Salaboy: And what differentiates Dapr and Diagrid from the competitors?

Bart: So, what differentiates Dapr is that it offers a suite of different Dapr Building Blocks that you can combine to build complex functionality. I haven't seen any other offering that provides all these building blocks that you can compose to build complex functionality. For me, that's a huge differentiation because you can start small by consuming one API, but then expand to consume more APIs that are already provided for you without adding any extra components or management on top of that solution.

Salaboy: What can we expect next from Dapr and Diagrid?

Bart: So, next, coming in Dapr 1.15, there are tons of things coming, but one important part is the Conversation API, which is an abstraction, an API that gives developers access to Large Language Models (LLMs). You can hook different LLMs behind the covers and let Dapr communicate with them via APIs. These APIs are smarter because they can cache returns, or responses from the LLM models, avoiding multiple hits to expensive services like OpenAI with the same query. This API acts as a gateway between the application and the LLMs, providing features such as obfuscation of sensitive data. For example, if you are sending emails or credit cards, you can wipe out sensitive information before sending it to the model, ensuring that the data sent to the model is safe for your company and does not expose sensitive information.