Einstein Platform Services

Basic Setup (Language and Vision need this)

1. First, install this repo into your org, and add the EinsteinApps perm set to users that need it.

2. From https://metamind.readme.io/docs/what-you-need-to-call-api#section-get-an-einstein-platform-services-account do the signup

   • you should end up with a .pem file

   • don't do the generate token section (the package does that for you)

   • If you've already done this once, you can't do it again with a user with the same email. You can use the same email and .pem file in multiple orgs.

   • we're going to upload this in SFDC. Keep this file handy and safe, in case you need to use it in future orgs.

3. Do the Upload Your Key section. Once the file uploads, be sure to share/make it visible to everyone in the org

4. In Salesforce, go to Custom Settings, to EinsteinVision, and then click Manage.

5. Create a new setting at the organizational default level

6. Set the Einstein Username you signed up with, and CertFile to be the name of the cert from step 3 (defaults to einstein_platform). Leave CertName blank (that's for people using the old signup process).

   • This username will be a real email address
   • If you used the oAuth flow from a Salesforce login to sign up (like the instructions), the username is the email address attached to that user, which may not be the same as the Salesforce username

7. Pick a userId to be the Einstein user (bonus points for having a user named Einstein and setting their chatter picture!). Make sure that user has the ChatterAs perm set.

8. token expiration time should be 3600

9. certificate issuer: developer.force.com

Validate the setup

1. Try the standard model: In Salesforce, create a vision model (you can call it anything) and set the Einstein Trained Model Id to GeneralImageClassifier
2. Post an image as a file or as a link to the Chatter feed (for files, use the Attachment paperclip in chatter, not the inline image [mountains icon] inside the post body)
3. Click on the comment box (live feed) or reload the page to see the response.
   • getting predictions and probabilities is good
   • getting "I don't have an answer for that" means your setup is probably wrong.

Goal of this project

• Access Einstien in Salesforce without writing code

Einstein Vision classifies photos.

Docs are here: https://metamind.readme.io/docs/introduction-to-the-einstein-predictive-vision-service

User creates a model and labels using Salesforce objects

• Users can control the probability threshold and whether secondary predictions are shown.
• Users can view (Lightning Experience Only) the model's statistics via a Lightning Component
• Users can access the auth tokens via the browser console

Users can clean up after themselves

• deleting a model in Salesforce deletes it from the PVS

Users get predictions

• Invoke the model by posting either a file or an image to the chatter feed for that Label.
• Einstein (via another user) will chatter back the predictions according to the model's settings.

Users can create examples via a local .bash script

	~/SomeOther/Place/directoryUploader.sh c7f7d45327504236c6dfd897cbeffd42a40be485 1952 795

where the three parameters are the token, the labelId, and the modelId. Note: you will have to make the script executable, like chmod u+x directoryUploader.sh.

After you create a model and some labels, click on a label to view the label-specific terminal command to run, including a valid token (#onlyInLightning). You're welcome!

This directoryUploader will take every image in the user's current directory and upload it to that indicated label. Notice that the script is in some other place, and the current directory only has the images in it.

Users can train a model

Just hit the train button on a model and the code will handle the rest #onlyInLightning Love those new Lightning Actions!

Custom Settings

• Manage your Einstein Credentials
• Manage the security cert used when you sign up for an account
• Be sure to populate these before trying to use the service

Using Standard Models

• For the food model, set the Einstein Trained Model Id to FoodImageClassifier
• For the general model, set the Einstein Trained Model Id to GeneralImageClassifier

Fun things to do

• be sure to walk around with Salesforce1 attaching photos to a model of anything around.
• Have Einstein idenfity dog breeds

===

Einstein Language

Follow all the Predictive Vision Setup stuff.

Invocable method from ProcessBuilder for calling standard or custom language models.

1. create at least a custom text field to store the prediction

2. create a process builder on the object. Pass in the required fields (recordId, field to analyze, and the field to store from step 1)

   • make sure the fields are not null
   • set up your criteria so that you're only requesting predictions when the field to analyze changes or is new. This is important--recursion is not your friend.
   • Example criteria for analyzing case subject

   and(
     not(ISBLANK([Case].Subject)),
     or (
       and (ISNEW()),
       ISCHANGED([Case].Subject)
     )
   )
   

3. optionally, store things like the probability (another custom field...number, not text. I used 3,3 for my decimal places), raw results (another custom field) or use a custom modelId+type instead of the default sentiment

4. Create some records! You'll have to refresh the page to see them, unless you're using the LiveRecord component/event (included!). More on that here: https://github.com/mshanemc/liveRecord

Invocable method from ProcessBuilder for supplying feedback (corrections to the model).

Lightning Components for creating custom language models based on

• data in Salesforce
• files attached via Chatter

Components for

• total Einstein Usage/Limits
• showing all labels and example count
• monitoring the status of model training completion

===

Debugging/Problems

I've got a lot of system.debug stuff running in my code. If you're seeing something not working, please open up the dev console, go to Logs, and then request your predictions. You'll see the logs coming in, and get some hints about what might be wrong.