Inspiration

In hospitality & travel, there are only three must buy services - the flight (or means to get to somewhere), the hotel (or other accommodation) and food.

The majority of mid to upper range hotels in EU / US have on-site dining. Lets work out how we can combine travel preferences and identity that will be used for the room reservation process and apply that approach to on-site dining.

Also wanted to dig into AI agents - will these be useful? Can they be a benefit to consumers or are they perhaps more about streamlined hotel operations. This hack was a good way to discover what the future may be for hotels and AI agents.

What it does

Start with two DIDs - one for the hotel (in our hack, issued by TBD), and one for the consumers (in our hack, issued by Autoura.me). For example the consumers DID might have been shared during the initial room reservation flow or at point of checkin.

First the hotel accesses the guest's preferences using the Autoura DID enabled preferences endpoint. https://www.autoura.com/docs/api/profiles/preferences

Then, using AI, these guest preferences are compared to the special dishes available at the on-site dining. For example if the guest has indicated they like burgers, and there are burgers on the menu, this is a match.

Finally, the hotel uses AI to create a message that is personalised for the guest highlighting the food that is available THAT THE GUEST can and will want to eat.

The hotel initiates communication to the guest, sending this message via DIDComm. (The Autoura platform handles consumer DIDComm messages, visible in the Autoura Connect app).

How we built it

Most of the focus for this hackathon was getting the core Autoura platform ready for this capability. (Core platform including DID issuing, Autoura Connect iOS/Android apps for DID management)

The code (in Github) is built in Vue / JavaScript. It uses OpenAI Chat Completions endpoint for the AI element. Uses Autoura.me for preferences. The code represents what the hotel needs to build.

Consumers can manage their preferences via the Autoura Connect iOS / Android apps.

Autoura consumer DIDs are time limited so for example expire after a guest has checked out of their hotel. This helps maintain consumer privacy plus also minimises any unwanted commercial promotions.

Challenges we ran into

DIDComm encryption was challenging to achieve in JavaScript. Mostly the problem is there is no independent tooling to check we have got this right. (It works within our platform fine, but have we approached encryption exactly the same way as everyone else?).

Accomplishments that we're proud of

Very pleased this works and seems to be something valuable that hotels would actually use (with a little more development time for extra polish).

Also I think we have worked out how to pitch decentralised preferences to regular consumers. The secret seems to be to use any word but "identity" ;)

What we learned

Learnt that ChatGPT is pretty good at some of the more complex encryption code, but can't engineer decentralised identity based approaches very well (likely because there isn't much information available in its training dataset). Impossible without it though, at least for me.

What's next for Personalised dining offers using AI & travel preferences

I wanted to build the consumer end of the message. e.g. the consumers AI agent receiving the message from the hotel and going, oh yes please book me in for a table.... or saying, I am going to the gym, I need a light meal only..... an entire two way conversation - AI agent to AI agent - in freetext.

We have the foundations for that now in our core platform but couldn't quite get it all to the level of polish necessary to be a public release..... a few more weeks ;)