Sous is a generative AI cooking assistant that helps home cooks build meals around the ingredients they already have - and stays out of the way when they don't need it.
At a glance
Role: Designer (team of 4) Methodology: Design Thinking Tools: Figma I owned the nutrition tracker feature end-to-end — user flow, mobile screens, tablet redesign — and led the design system rollout across the entire app. I co-led usability testing and shaped the interview protocol that grounded our early direction.
Why Sous
The team started from a shared irritation: staring into a full fridge and still not knowing what to make. Existing recipe apps solved a different problem - searching for something you already had in mind. Nobody was solving the messy reality of dinner planning - the ingredients on hand, the nutritional gaps that week, the energy you actually had to cook. Sous was our answer. A generative AI cooking assistant built around three ideas:
- Recipes suggested from your ingredients, not the algorithm's greatest hits
- Nutrition tracking that responds to what you actually ate, not what you meant to eat
- Cooking guidance that respects the user's autonomy - including hands-free voice interaction while cooking
The name came from what we wanted the AI to be: a sous chef, not the head chef. Someone who preps, suggests, and stays out of the way when you don't need them.
Approach: assistive, not authoritative
Most AI-first products default to doing more. Suggest more recipes, log more meals, generate more content. Design Thinking as our framework gave us a way to interrogate that instinct. We ran the full double diamond — empathize, define, ideate, prototype, test — but the questions the framework surfaced were about restraint, not capability. What should the AI do without asking? What should it never do without asking? Where does helpfulness cross into taking over? Interviews revealed the answer clearly. Users who cared about cooking — the ones we most wanted the tool to serve — kept describing the same fear: that a smart cooking app would flatten the parts of cooking they actually enjoyed. Discovering a recipe. Improvising with what's in the fridge. Deciding for themselves what "balanced" meant that week. The design tension became explicit: the AI's job is to give the cook more options, not fewer decisions. Every downstream choice came back to that principle.
What I designed
Nutrition tracker. I owned this feature end-to-end. The design problem was that most nutrition apps either infantilize (traffic lights, gamified guilt) or overwhelm (raw macros, no context). Neither served our user, who wanted awareness without judgment. The solution was to show nutrition in relationship to the week's cooking, not against a fixed daily target. What did you eat today, this week, this month? Where are your gaps? What ingredients in the fridge would fill them? The tracker feeds back into the meal suggestion engine - turning "what should I cook" into "what should I cook given what my body's been through this week." I designed the full user flow, ran think-aloud usability testing on the low-fi, iterated through feedback rounds, and developed the hi-fi in Figma. For the tablet redesign, I organized nutrients into semantic groups - macros, vitamins, minerals - with a horizontal stacked bar reserved for the macros view. This solved the "many nutrients in an uneven grid" problem before it became one.
Design system rollout. In our third sprint, we hit the standard mid-project chaos - screens designed by different teammates were drifting apart visually. I led the consolidation: locked the color palette, typography, spacing scale, and component library, then applied the system across every screen in the app - auth, chat assistant, recipe suggestions, nutrition tracker. Muted sage as the primary, warm cream as the surface, terracotta as the state accent. Fraunces Bold Italic for editorial serif moments, Inter for the rest. Bootstrap Icons for consistent iconography. Not glamorous work, but the kind of infrastructure design that separates a coherent product from a portfolio of screens.
Usability testing. I co-led two rounds of think-aloud testing on the low-fi and hi-fi prototypes. The framework was straightforward but disciplined: pick tasks that stressed the design's assumptions, observe without steering, note where users invented their own logic rather than following ours.
Where it landed
Sous was a coursework project - but the design questions it surfaced kept working on me afterward, especially the AI-autonomy tension and what "designed for context" really means. The Figma prototype is live and linked above.
Reflections
Detachment from your own work is a skill, not a feeling. Watching a user navigate the nutrition tracker in usability testing, I noticed myself wanting to jump in and explain. That impulse was the tell — if the design needed my narration, it wasn't ready. Sitting in that discomfort silently is one of the harder muscles to build as a designer.
AI doesn't remove design decisions, it multiplies them. Every capability an AI adds is a question the designer has to answer: does the user want this? Should this happen automatically or on request? What does the interface look like when the AI is wrong? Sous taught me that designing for generative AI is less about the model and more about the guardrails you build around it.
Systems thinking is design thinking. The design system I built in sprint three had more impact on the final product's coherence than any individual screen. Locking the palette, the type scale, and the component library was the decision that let every subsequent screen feel like part of one product rather than a collection of designers' opinions.