Build in AWS an AI-powered call center for restaurant chains, with human hand-off in case of any issues
Budget: $1,500 – $3,000 USD
Procedure and requirements for the call center:
1. The customer will dial the call center's phone #. The company has 4 brands, there is a single phone # and the AI voice agent will ask which brand the customer wants to order food from.
2. The call should be answered by an AI voice agent. The agent should have a natural conversation with the guest to take the food order, collect payment and inject the order into the POS.
2.1. Multiple customer calls may be received at the same time, so the AI voice agents needs to handle the workload, and answer within 1-3 rings.
2.2. The restaurant's API has all requirements: get today's promos and menu, lookup/create customers, get restaurant's hours and addresses, generate payment links, insert order into POS.
3. If the AI voice agent can't complete the order (whether due to lack of understanding, API or any other issue), the AI voice agent will:
3.1. Save to the 'web portal' a transcript of the call and the list of items in the cart.
3.2. forward the call to a secondary phone# where a team of humans will pickup and finalize the order.
Other requirements:
a. It is preferred that the AI voice agent is a fine-tuned model based on the restaurant's menu, ideally trained with real audio in order to recognize the menu items. I will be able to provide the audio samples, so you as freelancer can focus on fine-tuning.
a.1. As a 2nd preferred choice, a RAG technique may work as long as it can recognize utterings that are not actual/natural words, e.g. signature item names, promotion names, name combinations, or names that combine more than one language.
b. The call center's phone # needs to be 'ported' or configurable, as it will be a Latin American phone#.
c. 'Web portal' additional requirements (as mentioned above in 3.1)
c.1. Will need to store some details of all calls received through the main phone#. Needs to save the call transcript, timestamp, order status, order captured (as jsonas friendlier readable list).
c.2. As described above in 3.1, this portal will be mainly used by the team of human agents who will support the AI when it couldn't complete an order.
1. The customer will dial the call center's phone #. The company has 4 brands, there is a single phone # and the AI voice agent will ask which brand the customer wants to order food from.
2. The call should be answered by an AI voice agent. The agent should have a natural conversation with the guest to take the food order, collect payment and inject the order into the POS.
2.1. Multiple customer calls may be received at the same time, so the AI voice agents needs to handle the workload, and answer within 1-3 rings.
2.2. The restaurant's API has all requirements: get today's promos and menu, lookup/create customers, get restaurant's hours and addresses, generate payment links, insert order into POS.
3. If the AI voice agent can't complete the order (whether due to lack of understanding, API or any other issue), the AI voice agent will:
3.1. Save to the 'web portal' a transcript of the call and the list of items in the cart.
3.2. forward the call to a secondary phone# where a team of humans will pickup and finalize the order.
Other requirements:
a. It is preferred that the AI voice agent is a fine-tuned model based on the restaurant's menu, ideally trained with real audio in order to recognize the menu items. I will be able to provide the audio samples, so you as freelancer can focus on fine-tuning.
a.1. As a 2nd preferred choice, a RAG technique may work as long as it can recognize utterings that are not actual/natural words, e.g. signature item names, promotion names, name combinations, or names that combine more than one language.
b. The call center's phone # needs to be 'ported' or configurable, as it will be a Latin American phone#.
c. 'Web portal' additional requirements (as mentioned above in 3.1)
c.1. Will need to store some details of all calls received through the main phone#. Needs to save the call transcript, timestamp, order status, order captured (as jsonas friendlier readable list).
c.2. As described above in 3.1, this portal will be mainly used by the team of human agents who will support the AI when it couldn't complete an order.