When a referral intent is detected, the user will be given a referral link that they can share to others. Clicking on this link takes them to a small form to fill out their first and last name and their phone number. Users will be autopopulated into Airtable and Firebase will keep track of the referrals and who has used them. Later, referrals can be used to grant free lunches when a referred person gets lunch.
The orchestrator has been updated to have the following intents:
Chatting, used for general conversation and questions with Mountain Mama
Memory, used for recalling deep context information about the user
Functionality, used to fetch the documents from the Mountain Mama documentation
Location, used for setting and editing the delivery location of a user
Payments, used for setting and modifying a user’s payment method
Diet, used for setting/editing dietery preferences of a user and checking if a menu item falls into their diet
Order, used for placing and modifying orders for a user
Clarify, used to check the intent of a user if it is unclear or ambiguous
The orchestrator now scores each possible intent from a value between 0 and 10 and then determines the correct intent by picking the highest intent (No tiebreaker system has been implemented yet)
Multi-intent orchestration is planned
Long Term Memory Integration added
Long term memory stores key user interactions and useful tidbits for later recall. Currently used for just chatting intent and used to customize daily prompts. It is also used in the memory intent to recall the entire memory of a user.
Stripe Payments pushed to production
Prompt Managing system
Prompts now have a tagging system between ‘Technical’ and ‘Normal’ used to categorize whether they are front facing or backend system prompts
Prompts also now have a description associated with them and where they are used in the code for further clarify