AI Recipe Book: A Side Project About Trust

Why I Built This

I cook almost every day. It's my main creative outlet outside of work. The feedback loop is immediate, the materials are tangible, and nobody asks for stakeholder alignment on a pasta recipe.

Traditional recipe apps are frustrating. They can't answer "what if I don't have cilantro?" or "make this work in 30 minutes instead of an hour." Those are exactly the problems AI handles well: adaptation, substitution reasoning, contextual guidance.

I also had a professional reason: AI-powered features were becoming central to our product roadmap at Keela, and I wanted hands-on experience with conversational UI patterns before I had to lead design decisions about them.

How We Built It

I partnered with a PM and two engineers to take the product from zero to a functional MVP in under three months, shipping multiple iterations per week. Vercel for deployment, PostHog for real-time analytics, a modern serverless stack designed for experimentation.

As founding designer, I owned the end-to-end product experience: problem framing, concept validation, interaction design, AI workflow definition, and system-level UX decisions. I worked closely with engineering to translate complex AI behavior into clear, human-centered experiences, and introduced AI-assisted design workflows to accelerate iteration without sacrificing quality.

What I Designed

Conversational Adaptation

Instead of static recipe cards, users ask natural language questions: "Make this dairy-free," "I only have 30 minutes," "What can I substitute for shallots?" The AI responds with an adapted recipe and explains the reasoning behind each change.

The key insight: transparency is the trust mechanism. Users accepted AI suggestions only when they could see the why. "Use coconut cream instead of heavy cream" lands differently than a silent swap because it explains the reasoning for this sauce" lands differently than just swapping the ingredient silently.

This directly influenced how I later designed the automation preview feature at Keela. Same principle: make system logic visible to build confidence.

Step-by-Step Cooking Mode

A focused view that breaks recipes into discrete steps with timing. Each step shows only what you need right now: ingredients, technique, timer. Designed for the reality of cooking: messy hands, divided attention, a phone propped on the counter.

Instacart Integration

Users send a recipe's ingredient list directly to an Instacart cart. Bridging "I want to make this" and "I have everything I need to make this" turned out to be a high-value moment in the flow.

Shareable Collections

Users curate recipe collections and share them as link-based cookbooks. The experience should feel personal, like passing a handwritten recipe card to a friend, not forwarding a URL.

What Happened

• Zero to functional MVP in under three months, shipping multiple iterations per week
• Reddit post reached 120K+ people, driving early organic validation

Building end-to-end keeps me honest. I know what API latency feels like, what error states look like in reality (not just in a Figma frame), and where the gap between design intent and engineering reality tends to widen. That empathy makes me a better design leader.