The New Normal: Build the AI Agent You Wish Existed
From everyday frustration to a working grocery savings agent—exploring conversational UX, data integrity, and human-centered AI design.
The Challenge
Grocery prices vary from store to store, but figuring out where to shop for the best overall value is surprisingly difficult. Comparing individual products, quantities, brands, and store locations takes time—and most shoppers don’t want to visit multiple websites just to plan a grocery trip.
I saw an opportunity to simplify that experience with an AI-powered shopping agent.
What if you could simply tell an agent what groceries you need and let it do the comparison work for you?
The Solution
I designed and built Nourish, a mobile-first grocery savings agent that transforms a natural-language shopping list into an estimated price comparison across nearby grocery stores.
Instead of searching for products individually, shoppers can type or speak their grocery list, review the agent’s interpretation, choose their location, and compare estimated basket totals.
The goal was simple: reduce the effort between knowing what you need and deciding where to shop.
Designing the Experience
| 01. Start With a Conversation | Shoppers type or speak a grocery list naturally. Nourish interprets items, quantities, and brand preferences without requiring manual product searches. |
| 02. Keep Shoppers in Control | Editable quantities, list review, and optional brand substitutions ensure the agent assists without making unwanted decisions. |
| 03. Add Real-World Context | Google Places identifies actual nearby grocery stores using location or ZIP code, with adjustable search distances. |
| 04. Turn Data Into Decisions | Nourish compares estimated full-basket totals, highlights the best value, and provides store details and directions. |
The Data Integrity Challenge
One of the most interesting design challenges emerged during development: What happens when the agent doesn’t have complete or reliable data?
Early versions could encounter products without pricing information, resulting in incomplete comparisons and misleading $0.00 totals.
My original plan was to integrate real-time grocery pricing through Instacart. However, new developer access was not available when I explored the integration. Without access to verified retailer pricing, I needed to rethink how Nourish could still demonstrate meaningful grocery comparisons without misleading shoppers.
This wasn’t simply a visual design problem. It was a data quality and decision-making problem.
I redesigned the experience to use consistent, deterministic estimates when demo pricing data was unavailable, while clearly communicating that prices and inventory are not verified.
Nourish distinguishes between real store locations and simulated price estimates. That distinction is essential: a trustworthy AI experience should help users understand what the system knows, what it estimates, and where its limitations begin.
From Design to Working Product
I wanted this project to go beyond a traditional UX prototype.
Using AI-assisted development tools, I translated my product requirements, conversational flows, interface designs, and usability decisions into a working React and TypeScript application.
The experience integrates natural-language list processing, product data enrichment, location services, Google Places, estimated basket pricing, and interactive shopping trip details.
I used Kiro to accelerate implementation, GitHub for version control, and Netlify to deploy the application publicly.
Throughout development, I tested real shopping scenarios, identified breakdowns, and refined the experience—from handling ambiguous quantities to improving incomplete price comparisons.
What I Learned
Building Nourish reinforced something I’ve learned throughout my work in conversational AI:
A successful AI experience isn’t just about understanding what someone says. It’s about helping them make a decision they can trust.
That requires more than a polished interface or a natural conversation. It requires thoughtful handling of incomplete information, clear communication of uncertainty, and opportunities for people to stay in control.
AI-assisted development made it possible to bring the idea to life quickly. Human-centered design helped determine what the experience needed to do—and how it should behave when the data wasn’t perfect.
.