An AI agent that finds healthy groceries nearby
Concept
2026
While there are many apps the deliver food straight to your door, there are few that solve the real problem of, “Where should I get food?” This question is more difficult for individuals living in food deserts. food deserts are characterized by a lack of nearby supermarket, they are also characterized by chains of fast food stores that don’t offer healthy options of local residents, thus increasing health problems. Therefore, the problem is not ordering food, but rather finding healthy alternatives.
I created a multimodal voice AI agent that finds different food alternatives from supermarkets and farmer’s markets for those living in food deserts. The agent takes into account the person’s location, health goals, and economic circumstances to provide them with a list of locations where they can order and receive healthy foods.
This was created for Google’s DeepMind hackathon. But, after the hackathon I tested the app among 5 different users and noted a high user satisfaction in finding good quality grocery options.
The truth of the problem was double-sided. Low income minorities trapped in food deserts suffer the effects of being deprived of healthy food options while farmers suffer losses of unwanted produce. Therefore, there was an opportunity to resolve both user issues. app
PAIN POINT: The typical journey starts on delivery not on sourcing food
Key Insight: The app shouldn’t be just another food delivery application. There are already many app that accept EBT//SNAP. These apps require users to already know which stores carry what they need while farmers continue to remain invisible.
Market research
Although there were many apps that allow for food deliveries, there were few who were able to address and present a solution for both sides of the problem. While they delivered food, they did not address user’s health concerns nor did it they place farmers front and center.
Design exploration sketches
I prioritized a mobile first design to make the app available to more users
For the majority of users within my targeted demographics using mobile phones is practical, convenient and is on the rise. Therefore I prioritized a mobile first approach. However, I found that my initial navigation was too complex and could create confusion.
Throughout the hackathon, I started to narrow down my design solution to multimodal voice activated AI agent. An AI first approach eliminated the constant back and forth of different screens, but also allowed users to directly communicate their concerns without assuming what they might need. It also creates room for personalized experiences.
[insert graph of on-boarding user flow]
Taking data from users and adjusting with each search and scan
The agent still needed information from the users, and the on-boarding process was the best place to gather that information.
Crafting personalized experiences.
The search process was not the only feature of this app (method of searching). Fridge scans gave the AI more context about the user’s shopping patterns and health goals, thus allowing the agent to give more accurate and personalized results.
The MVP has been deployed via Vercel. To access, click here.
For camera access and microphone access, explore the version in Google's AI studio.
Although this app was created for a hackathon, I test this app with couple of friends. They were not part of the target demographics but the app still efficiently was able to provide accurate results and facilitate their grocery shopping experience. What I realized is that even though I was designing a solution for a very specific group of people, my solution could help anyone.
