WineWiz.png

WineWiz

WineWiz | An AI-Powered Wine Gifting Assistant

Web Application Design | Apr. 2024 - May. 2024 | Microsoft Generative AI Hackathon
Skill:
Product Management, Product Strategy, UX/UI Design, Prototyping
Release: POC(Proof of Concept): May 06, 2024
More Information: https://devpost.com/software/winewiz
Demo Website: https://winewiz.psyduck.ai/


Inspiration

Our inspiration for this project stemmed from firsthand challenges encountered in gifting wine:

  • Selecting an appropriate wine typically involves a laborious and unclear process, where consumers must undertake extensive research without definitive guidelines for choice. This difficulty is compounded by the scarce availability of expert consultants in wine shops, who are crucial for informed decision-making.

  • Customizing the wine gifting experience adds complexity, necessitating a personalized gift card that captures the unique relationship between the giver and recipient, along with specific details that underscore the particular importance of the selected wine to the recipient.


What it does

Wine Gifting Agent
Winewiz simplifies selecting and gifting wines by offering AI-powered conversational consultations and personalized gift card creation in a single mobile web app.

Key User Flow
After I defined the two key problems to solve: selecting an appropriate wine and customizing the wine gifting experience, I designed the key user flow that collects the user’s preferences through a guided conversation that is intuitive, low-effort and engaging. Wiz picks the perfect wine for the user and then provides a gift card template for users to upload and stylize their own photo to include in the gift card.


Chat About User’s Needs to Pick the Perfect Wine
The process of collecting user’s preferences is highly conversastional and engaging. WineWiz first asks about the user’s budget, then guides a few rounds of free conversation to effectively find out the user’s need. Throughout the conversation, WineWiz maintains a set of keywords that represent user needs, which will be used as search keys to search within our database.

Make a Gift Card
After the wine is picked, the user can simply tap on the purchase button to go to TotalWine to purchase it. WineWiz then provides the feature that generates personalized a giftcard that consists of a stylized personal image and a lovely paragraph according to the information collected in prevous conversations. The user can simply upload an image of his/her own and choose a style to apply. Our AI will stylize the image to include in the gift card.

Color Scheme
I crafted this color palette to complement a wine-themed aesthetic, enhancing the overall immersive experience.


How we built it

  1. Data Gathering: First, we developed a web crawler to gather publicly available wine data, subsequently refining its structure and content manually to enhance usability and relevance.

  2. Solution Crafting: Second, leveraging insights from extensive user research, we crafted a multimodal solution tailored for wine gifting.

  3. Technical Implementation: For implementation, we used Next.js based on the React library to develop the User Interface and ultimately deployed it on Azure. We utilized a suite of Next.js and OpenAI APIs. These APIs include Azure AI Speech-to-Text Service and OpenAI Whisper Text-to-Speech for interactive communication, GPT-4 Turbo for keyword refinement and text generation, an OpenAI Assistant for intelligent wine searching, and DALL-E 3 for generating custom gift card images.


Challenges we ran into

Back-end Development: We faced challenges ensuring the backend produced stable and consistent outputs via the OpenAI Assistant. Achieving reliable structured data output was crucial for seamless integration with front-end processes.

Content Creation: During the content creation phase, generating gift card images that met our aesthetic and thematic requirements proved challenging. The process involved meticulous extraction of image characteristic tags, strict adherence to thematic consistency with the chosen wine, and creating images in predetermined styles. We engaged in several rounds of iteration to perfect the prompts for image tag extraction and image styling.

Front-end Design & Development: In front-end design, our goal was to develop an animation that not only accurately reflects the system's status but also enhances the user's conversational experience. The primary challenge lay in detecting various system states and dynamically adapting the wave-shaped animation to these changes, requiring a delicate balance of sophisticated graphic design and precise status detection.


Accomplishments that we're proud of

Backend Development: We refined our backend processes through advanced prompt engineering, establishing a stable and reliable conversational experience. This refinement included the seamless integration of the "File Search" function from the OpenAI Assistant, while consistently delivering structured JSON data to the front end.

Content Creation: After numerous iterations to fine-tune our approach, we achieved precise extraction of image characteristic tags and adherence to thematic elements relevant to the selected wine. We successfully generated images in specified styles, enhancing the overall aesthetic and relevance of the content.

Frontend Design: We innovatively designed and developed a dynamic wave-shaped animation on the front end. This animation changes its form and amplitude based on different system statuses, such as listening, speaking, or thinking. The creation of this feature required an intricate combination of sophisticated graphic design and accurate system status detection.


Takeaways

Research Insights: Our research, which included insights from customers and Totalwine consultants, alongside the capabilities of Generative AI, revealed that a multimodal wine-gifting experience can be effectively enhanced by technologies like GPT and DALL-E. This approach offers a more dynamic and engaging interaction for users.

Technical Development: From a technical perspective, we developed an innovative framework for the wine selection process that surpasses traditional search functionalities in segmented markets. This new method proves particularly advantageous when users lack precise search terms or criteria. Despite its effectiveness, the existing "File Search" feature leveraged through the OpenAI assistant is not cost-efficient. Consequently, we are contemplating the implementation of a backend wine search feature that operates independently of OpenAI's token-based system. This alternative would enable broader scalability and database management without the constraints imposed by file size and quantity limitations in the current OpenAI model.


What's next for Winewiz

We launched Winewiz as a mobile web application in mid-April 2024 and are now poised to broaden its reach. Moving forward, our plan involves expanding the application's availability across multiple platforms, including web, iOS, and Android, to ensure a seamless user experience regardless of device. Additionally, we are actively exploring potential partnerships with local wine retailers and distributors. These collaborations aim to extend our business scope, enhance our data capabilities, and effectively utilize AI to support a variety of modalities, thereby enriching the wine selection and gifting experience.


Built with

Azure | Next.js | OpenAI-APIs | STT (OpenAI-Whisper) | TTS (Azure-AI-Services)


Try it out

winewiz.psyduck.ai

01-v2.png