AI powered sommelier
UX & conversational Designer, User Researcher
Project brief
The application was presented as a Proof of Concept (PoC) to the AI Summit 2023 for the integration of sentiment analysis in combination with generative text (OpenAI), Voice command using Natural Language (Alexa) and generated images (Dall-E) with the objective to demonstrate the capabilities of a human like interaction on wine recommendations.
Methodology
The first step was to produce a natural customer journey which would attract users.
The second step was to ensure that the integration of each of the individual AI components was properly placed within the journey and that they provided the expected result.
The last yet important step was to define a way of validating the human like quality of the entire experience with users.
Inception
This project was executed in a start up mode. As a UX & conversational designer my responsibilities included to assist the team of Front-end, Back-end and Data Analyst to ensure the validity of the journey. As part of the testing team, I also created scenarios and conducted the tests. With the help of the marketing department, we did a competitive analysis in order to identify the best PoC.
Through brainstorming sessions we opted for a sommelier application which would allow:
-
both generative text and generative image to be produced collaboratively
-
implement the sentiment analysis on the recommendation journey
-
produce a sensation of tailored proposition for the user's perspective
Provisional customer journey
Conception
Once the provisional journey identified, we proceeded to identify the best positioning for each of the AI elements. I then proceeded into building a conversational journey using edge cases to test the model.
User Flow for User, AI and device
Once the final journey designed, I then proceeded into building a conversational journeys using edge cases to test the model. The database was amended within multiple iterations to produce the expected result.
Example of a conversation using OpenAI
Testing
We produce a task plan with specific scenarios we wanted to test and questions to ask the users so we could validate our hypothesis. We concentrated on qualitative testing as the adoption by users was deemed extremely important.
I then compiled all the results and findings in a debrief document which would be used to further iterate on the model. The main finding was an awareness of the AI in general with some users identifying generative text.
Qualitative testing results
Interaction Design
The first contact with the user was key to the success of the interaction. We decided to use a multiple devices to enhance the experience. This included an Alexa tablet used for the voice command of the user and resulted recommendation and a large screen monitor to display the generated image
Display station for Sommelier PoC