Architecture
This content is for Spring ’26. Switch to the latest version for up-to-date documentation.
Overview
Section titled “Overview”Aforza’s Ava is built on a trusted and scalable cloud platform architecture that can meet the stringent performance and security demands of enterprise businesses. Ava uses cutting edge technologies and generative LLM models to create a best in class Vertical AI solution for Consumer Goods companies. The Ava architecture can broadly be summarised across 3 primary layers of technology.
User Applications
Section titled “User Applications”Ava is accessed through a number of purpose built mobile and web applications, meaning you can use Ava wherever you are and on whatever device you want. Ava works as a standalone application or can be embedded within your existing Aforza applications. Ava’s mobile applications are built using ReactNative and work on both iOS and Android operating systems. Ava’s web applications are cloud based, built using React, and can be accessed on any modern browser.
LLM Microservice
Section titled “LLM Microservice”All Ava based applications are routed through Aforza’s LLM Microservice which coordinates the entire conversation process. The LLM Microservice is hosted on Aforza’s GCP application stack and consists of robust authentication, secure APIs, and scalable storage architecture. Aforza’s Microservices are built as containerized Node.js applications within Google Kubernetes Engine. The LLM Microservice is incredibly performant and scalable, able to process millions of conversations. When using Ava, no matter what channel, the conversation is handled by the Aforza LLM Microservice, making sure that the context, data, and user information is properly routed and handled by the correct Agent and LLM Model.
Vertical AI Models
Section titled “Vertical AI Models”The model layer of the application stack is where the Generative AI model processes the request and generates the appropriate response. This response might be text based, or it might be a notification to call a specific tool to gather more information. Ava is built in such a way to be model and provider agnostic, meaning that over time our customers can choose between a variety of providers to meet their speed, cost, and use case requirements. Aforza Ava initially runs using models on OpenAI’s platform, but this will expand with time to include other vendors.
These 3 layers provide a flexible and powerful way of handling conversational requests to the Ava platform. The rest of the architecture section goes into further details of some key areas and concepts that sit within these layers.