Intelligent Photo Service (IPS)Master Project Part 4
System Architecture
Salesforce Is the Leading cloud-based company and provides SaaS and PaaS to many enterprises. Everything offered by Salesforce resides in multi tenant trusted cloud. Salesforce stores customer data, provides process to nurture customers, gives ways to communicate and collaborate with people you work with.
Intelligent photo service application delivers highly customized user interface to customers, employees, and partners. The application architecture consists of series of layers that placed on top of each other’s.
The application sits on top of the platform. It offers two communities — Photographer community and client community both are resides in community cloud. Communities are online social platform for users of an application. The important part is, Einstein model communicates with both communities to handle and manage user’s data.
The most important layer is Data Model. the application data model resides here. The data and metadata stored in Salesforce standard objects like Accounts, Contacts, cases, and custom objects namely, Photographer, Client, Album, Event type, Photo Collection.
The next layer is Platform micro services which includes many declarative approaches of an application like, Assignment rules, Flow Builder, validation rules, workflow processes, mobility and many more. These micro services are tiny and broadly decoupled little tasks to use in complex processes.
Einstein Vision API
Einstein Vision is a crucial part of Einstein platform technology services. IPS utilizes the functionality of Einstein vision API to AI-enable the application. Einstein Vision provides many pre-built classifiers or users can build custom classifiers to concur the image recognition use cases. IPS incorporated the power of image recognition to IPS CRM and enhancing the functionality of an application.
The main components of EV — AI are Datasets, Label, Model, Training and Prediction.
Datasets
The trained data includes input and output and generate the model which will use to make predictions.
Label
A label references the output name which will predicted by the model and input name is the group of similar data.
Model
Machine Learning developers create and train the dataset to predict the respected outcome. When developer train the dataset, the system will determine the similarities and differences between the multiple labels to categorize and characterize to define the label.
Training
It is a process in which model is created and learns the classification rules from given dataset.
Prediction
The outcome generated by system is useful to identify how close the input data matches to the predicted result.
Machine Learning Life Cycle [MLOps]
Intelligent Photo service application follows MLLC periodic or recurring process which involves Plan, Package, Create, Verify, Train, Configure and Monitor Phases.
There are six phases of Machine Learning life cycle (MLLC).
Plan
The plan phase includes the process of defining the event types which are used to rank the photographers. Its IPS responsibility to update the event type list regularly or based on requirements. Currently Event type list consists of Party Images, Wedding images and sports images.
Package
After finalizing Data or event images in plan phase, its required to create a package. This phase includes name the images according to standards, arranging images in proper format and finally convert folder or package into zip file. [Salesforce. (n.d.). Einstein ai. Retrieved from Salesforce Einstein ai: https://einstein.ai/research]
Create
This phase is crucial as it allows to generate the valid Zip URL of the dataset folder. It’s necessary to provide the path to the zip file. To generate standard URL and provide the access from external hosted site into Salesforce is the most important stage of the life cycle.
IPS event list zip folder hosted on GitHub and used later in Einstein Vision Custom Component located at Lightning Welcome page of an application.
Verify
After Successfully finishing the Create phase, the administrator needs to click the Create Dataset button. If zip URL format meets the standard of salesforce, the dataset will be created with his name.
IPS trained the EventList dataset which includes twenty images of Wedding, twenty images of party and twenty images of Sports. After inserting zip URL and clicking create dataset button, the dataset will be created, and later business administrator will train the dataset by clicking Train button from Intelligent Photo Service Welcome lightning page. It is the responsibility of IPS to continuously create and update the dataset and trained them.
Configure
In IPS, salesforce developer and administrators can create and trained multiple datasets but only one dataset will be active at a time. Currently, four datasets are generated and trained successfully. For this IPS application, EventList Dataset activated from Custom Settings of salesforce setup. It means currently, default organizational level value is name of EventList dataset.
Monitor
Its IPS administrator and developer’s responsibility to continuously monitor dataset and MLLC. If client or photographer requested to update EventList, it will be updated in picklist of data model and the respected dataset folder. Business administrator will rank the photographers daily to update the rank as well as to monitor that whether the ranking is updated for specific event type.
Next Part of this story
