Adeptia AI Services
- Rajeev Rai
Table Of Contents:
Purpose
This document aims to deliver a comprehensive overview of AI Services Deployment. It outlines a thorough process that includes procedures for Infrastructure Deployment and Configuration using AC5.x Professional.
Scope
This process encompasses the deployment of AI Services within the production environment and its integration with AC 5.x Professional.
Infrastructure Deployment in Azure
1. Create a “Resource Group” with required “Tags”
Create “App Service Plan” with required “Tags”
Once “App Service Plan” is created, it will show the overview as below:
Create “Web App” with the steps below:
Put name, Publish Type, Runtime Stack, Java Web Server Stack, OS and Region
Select Pricing Plan, as per recommendation from Product Team
Give reference from where the image will be pulled
Select Networking Options, keep public access as “ON” and Network Injection as “OFF”
Select “Monitoring” option as follow for Application Insight
Give the Tag
Once a Web App is created it will show like this, we can hit the URL to test the default page.
Add environment vairables in app service
Deploy through pipeline - Build Pipeline
Release Pipeline
Create “API Management Service”.
Configure API’s
Add required policy
Update backend
Create GET & PUT operation
App Reg and integration
Create new App reg
Configure certificate and secret
Add the client id in previous created application
Add JWT Token in API management
Update values according to the Application created
Create Log analytics Workspace
Create application Insight
Integrated application insight into api management
Add the required configuration to send logs to application insight
Update the inbound policy for extract unstructured data async and change the client id according to the created details
Create AKS Cluster
Configure Milvus
attu:
enabled: true
service:
type: LoadBalancer
extraConfigFiles:
user.yaml: |+
common:
security:
authorizationEnabled: true
service:
type: LoadBalancer
minio:
persistence:
size: 50Gi
pulsar:
bookkeeper:
volumes:
journal:
size: 50Gi
ledgers:
size: 50Gi
kafka:
persistence:
size: 50Gi
Default username and password
Username - root
Password- jpQfSvbKUkZG
Create new user and password to use in application
Also change the password of root user
Login using Attu loadbalancer:
Create new database - Database name should be start with underscore "_" with client key , Like "_<<Clientkey>>" where all the dash"-" needed to be replaced by underscore "_"
Here: Client key / id = The app reg client id
To onboard new clients we need to add new clients in app reg and also create the database of the same.
Configure AI Service
Go to manage deployment
Create new model deployment
Increase the token limit to max for all
Create Content filter
Requested decreased restriction on open ai content filtering
Request Quota increase:
Document Intelligence
Create document intelligence service
Go to document intelligence Studio
Use postman to migrate studio from sandbox to production
https://drive.google.com/file/d/1gcq2Fp0QrfMHAG_gCmPi0cBcH-oJ7c1E/view?usp=drive_link
Get the Model after migration using below
Document Models - List Models - REST API (Azure Azure AI Services)
Security:
API Manamagent: Only allow IP address of AC5 professional AKS cluster in api management
App Service : Only allow request from api management
Document Intelligence: Only allow request from with the Vnet
OpenAI : Only allow the request from within the Vnet
Patching and Upgrade:
Below process will be used for application upgrade
Add slot in prod
Update version / patch in different slot
Swap the prod with prod-02