Inspiration
The project stemmed from conversations with a sommelier friend and a wine shop owner. They both highlighted the potential of chatbots in-store (place QR codes around shelves) to assist customers in selecting wines, especially during busy times when staff are limited.
What it does
Hourglass is a conversational AI agent designed for wine shops, acting as a sommelier for in-store inventory. It enhances the customer experience by providing expert wine recommendations and answering inquiries in a personalized, efficient, and friendly manner.
Please note that the database is currently limited to a nearby wine shop with an inventory of 50 wines. The idea is it should be hyper-specific to shop.
How I built it
The agent utilizes three main tools:
- Wine Inquiries: Built using OpenAPI, Flask, and Google Cloud Functions for general wine-related questions.
- Inventory Search: Uses an unstructured product catalog within a Bucket/datastore to search inventory descriptions based on flavor profiles, tasting notes, or food pairings.
- Price Check: Retrieves the current price of specific bottles of wine by searching the inventory database using OpenAPI, Flask, and Google Cloud Functions.
Challenges I ran into
Ensuring the accuracy of wine recommendations required examples and refining recommendation algorithm.
Accomplishments that I'm proud of
Receiving positive feedback from sommeliers and wine shop owners during the testing phase!
What I learned
I found it remarkably straightforward to build an agent using the Google ecosystem. The vertical integration of Google’s data stores, embeddings, models, agents, and cloud functions allowed me to build the entire pipeline, create examples, and implement a UI within one platform.
What's next for Hourglass
The plan is to expand Hourglass from the initial testing phase with the sommelier and wine shop to other wine shops and specialty retailers.
Log in or sign up for Devpost to join the conversation.