top of page
final.png
Writer's pictureMad Scientist

Supporting Developers on Azara


Cognitive Architecture - Where we play

Hi, I'm Steve aka madscientist, CTO of Azara,


Welcome to my technical series of posts which will introduce you to some of the current and upcoming features of our Azara.ai platform. 


This series will cover the following topics:

  • Introduction to Azara from a developers perspective,

  • Technical architecture overview

  • Core components: Plugins, Channels, Workflows, Agentic Scenarios

  • Extending Azara.ai using SDK, Developer Tooling, Plugins & Agentic Scenarios

  • How we use AWS Bedrock to generate Agents, and optimize prompts

  • Deep dive into how we use Langchain & Langgraph in the platform.


If you are keen to try out the dev features of Azara, you can sign up for the SDK release here

With that, lets dive in and explain what Azara is about.

About Azara

Azara is a no-code platform which enables non-technical users to be able to create sophisticated agents to chat, do document querying, interface to multiple external systems, and create and manage workflows for work automation. It does this by allowing developers to easily create plugins for 3rd party applications, being able to upload agentic scenario code to extend agent behaviours, and making the RAG process for Large Language Models (LLM’s) trivially easy.


For the layperson, this means that in order to create a sophisticated agent that can work on your behalf, all you need to do is:

  • Create a new agent - we can even generate that for you.

  • Upload documents and websites to scrape to your agents’ knowledge base

  • Test your agents abilities by using the Chat preview

  • Deploy your agent to a Web Widget (WW) or to a channel (e.g. Whatsapp, Slack, Email, Asterisk/Telephony  etc.)

  • Extend the systems that your agent can query and manage by enabling an existing plugin (e.g. Hubspot, GMail, Calendar, etc) 

  • Create custom forms to collect information from your users via the WW or other channels.

  • Generate workflows to automate tasks and process information from user data collected in forms, or on a schedule.


It does all this in a beautifully presented and fast, optimized system that has a high focus on user experience.


How Azara Compares to ChatGPT


Where do developers fit into this?

So where do developers fit into this no-code solution then? 


A beautiful, optimized UI & UX for your users

Well, there will always be a need to keep extending the capabilities of this platform, and rather than the Azara team creating all the integration plugins, which quickly becomes a bottleneck, we will allow developers to use our platform to build on, extend and deploy their own solutions to their customers.


The advantages for them is that they get a highly optimized stack, which removes the need for them to tune, or write customer facing UI’s, they can focus on the business logic for their specific use case.


Azara is a Platform for doing agentic development, which just happens to have 

  • A beautiful UI,

  • No-code interface for your business users,

  • Multi-LLM support tuned for performance & streaming, 

  • Plugins & Channels for easily extending access to 3rd party systems,

  • Workflows to automate processes using the plugins,

  • Prompt optimization built in, using fine tuning and/or distillation,

  • Agentic Scenarios to extend the agent logic, or create Multi-agent implementations


Azara.ai Web Widget (WW) in action.

Azara Web Widget

Extending Azara

There are 5 main areas where extending an LLM Agent application can be painful:

  • Plugins: Integrating 3rd party services, e.g. Hubspot, Calendar, etc.

  • Channels: Adding additional chat channels to interact with agents, e.g. Slack, Whatsapp, Telephony, etc.

  • Agentic Scenarios: Extending capabilities of agents for specific Scenarios, e.g. Agentic RAG (Multi-hop queries, or checking for hallucinations etc), or use case specific logic.

  • Workflows: Ability to allow the non-technical users and domain experts to create workflows which run cheaply at scale without hallucinations

  • Prompt Training & Optimization: Ability to generate Agent Prompts and optimize them against their knowledge bases using either fine-tuning or distillation


By making the Azara.ai platform a composable architecture, we can easily extend its capabilities without having to be tied to the development cadence of the core server, which becomes important as the number of users on the system scales.


SDK for automating the Azara platform

Azara.ai provides a developer SDK in various languages, e.g. Python, javascript/typescript, Go.

This allows you to integrate the functionality of the platform into your existing applications, developed in the LLM framework of your choice, e.g. Langchain, Llama Index, DSPy, etc.


The SDK will give secure access to the platform via an issued API key, and your customers will be able to login/test your application, purchase credits or upgrade features via your reseller portal.


Future Marketplace

As demand grows, we will consider allowing developers to allow their plugins, channels, and agentic scenarios to be used by other developers and customers.


Note that for private use, this step isn’t needed, and plugins, channels, and scenarios are available by simply hot loading them from the UI or via the SDK.


In the next blog post, we will dive into the technical architecture of the platform, before covering the specifics of implementing the core composable architecture plugins.



Till next time …


Contact:

@steve_messina

@ai_azara

madscientist


Comments


bottom of page