Sous AI

Creating an AI-native, AI-powered mobile app that brings joy back into the kitchen, allows users to make quick decisions, tracks ingredients, and reduces waste

Summary

Sous AI started with the goal of leveraging AI to provide anyone with the tool that supports their lifestyle in and out of the kitchen. The mental load and skill gap when it comes to cooking can easily keep people from cooking - I know because I’ve been there. Using an AI-native approach, I set off to challenge myself to think about how I can create the right tool.

Role
Product designer, end-to-end design

Duration
1 week to MVP

Tools
ChatGPT, v0, Figma Make, NotebookLM

Product type
Mobile app

The goal

Utilizing AI tools to produce a MVP prototype that includes:
:

  • Smooth user flows

  • Acceptable UI

  • Multi-modal inputs and/or AI models

Why are we doing this?

  • Create simple and accessible solutions that are shippable

  • To test and push the limits of AI

The process

Discovery

  • User stories

  • User flows

  • Competitor Research

Converging

  • Opportunities

  • Deliverables

Ideation

  • Exploration and ideation

  • Iteration

Results

  • Final MVP designs

Initial insights

I turned to ChatGPT to be my partner and help me think through the kind of mobile app I can design. Together we explored different AI powered features and models and I decided to utilize Gen AI and visual and voice recognition.

Asking ChatGPT to use deep research led me to uncover 3 key pain points:

1. Ingredient recognition

Users have shared frustration with misidentified items

2. Recipe accuracy

Users were experiencing vague measurements, inconsistent quality, etc.

3. Waste tracking

This feature is rare, but could be very helpful!

I was not too surprised by all the competitors out on market as I’m sure the challenges of finding time to cook during a busy day is are widely prevalent. So I expected to see products similar to what I was envisioning. However, these insights validated that my product could have a unique position in the market.

Key insight: For a fun side-quest, I used Google’s Notebook LM to further test mark positioning and see if there are any potential benefits or pitfalls of this product. There wasn’t anything new noted, but heard some further validations!

The driving question became clearer:

How might we create an intuitive and personalized experience for tracking ingredients, cooking, and reducing waste?

Using the information I gained through GPT, I created a simple end-to-end user flow to see how and where the app could use gen-AI and voice and image recognition. Turning back to ChatGPT, I asked it to create user personas of three potential end users I could keep in mind while designing the app, and to generate a prompt for v0 to jump right into prototyping!

Rapid prototyping

The first prompt output was honestly not that great. Version 1 was very basic and minimal, and even with a few more prompts to make edits, there was still more work to be done.

version 1

version 2

Toggling between inputting my own prompts in v0 and going to ChatGPT to help generate prompts, I wasn’t getting the results I wanted and it felt like a slow, tedious process. So I decided to pivot. I took my designs to Figma Make and continued to refine my designs there.

Through Figma Make, I was able to develop more on the UI, task flows ad organization to where I felt it was ready for some feedback.

I asked ChatGPT to review our established personas, and by embodying their perspectives, to review and critique the designs. Additionally, I asked three peers for their feedback via short usability tests asking the following prompts:

  1. Input ingredients using text and generate recipes

  2. Input ingredients using voice and generate recipes

  3. Input ingredients using imaging and generate recipes

  4. Where would you go to find generated recipes?

version 3

Key insights: through GPT and user feedback, the overall strengths were identified as:

  • Thoughtful feature parity across all input methods

  • Easy flows

  • Clean and straightforward

  • Good feedback cues

However, there were still room for improvements such as:

  • Adding more filters and tags

  • Expanding on post-recipe engagement like, saving, rating, etc.

  • Improving flows to make it feel less “clunky”

Current designs

The user feedback had me thinking, especially about how I could further develop the overall input experience. After brainstorming with GPT, I decided to make the app really feel more like a cooking-enthusiast friend or a sous chef by creating a chat that would house all input options. All while still prioritizing an intuitive and personalized experience.

Home tab:

  • A personalizing the home tab to make it feel more inviting

  • Simple interface makes it easy for user to see what quick actions they can take

Input ingredients and get recipes:

  • The main function is a chat that supports multi-modal input of ingredients (text, voice, or image), and generates recipes in return

  • Chat is to give users the feeling of a helpful friend

Encouraging less food waste:

  • Using icons, tags and hint text to let users know what to cook and eat soon on recipe cards and ingredients listed in the home tab and full recipe

Impact:

  • Significantly made workflow quicker and more efficient

  • Increased opportunities for collaboration, testing, and feedback

Next steps:

  • Refine the chat function

  • Expand on post-recipe engagement like, saving, rating, etc.

  • Enable ability to log and save conversations

Key learnings:

  • AI is a friend, not foe!

  • Focus on the Minimum Lovable Product to ship and gather feedback quickly

  • There are details when designing for AI-powered features that I learned so much about

Go to the next project:
SaaS for Clarity Media Partners