Brian Mohr · Chicago

Product Designer
who ships.

My skill set is unique. I design and build.
I've been the founding designer at three software companies, always embedded with developers.

Foodbe AI brain

Case study — Foodbe

Food content is broken. Nutrition advice contradicts itself. All the data is the same.

Foodbe is an AI food intelligence engine that reasons, resolves, and never guesses. It runs on a Gated Truth Architecture, where every response is generated from a curated, verified fact library. If it doesn't know the answer, it says so. Every failure is tracked and a new fact is added to the library. Because it goes deeper than data, into the science, culture, and technique behind each food, it gives users insights they've never seen before, surfaced at the exact moment decisions get made.

Download on the App Store

The Problem

Food content online is broken. Recipe sites are SEO-bloated. Nutrition advice is contradictory. The current AI landscape retrieves truth from sources that may or may not be correct. GPT, Claude, and similar tools disclaim their errors and walk away. The user is left to fact-check answers that sound authoritative but aren't.

I wanted to build something different: an AI platform that manufactures verified truth instead of retrieving it. Every answer grounded in hand-curated knowledge units, anchors, citations, scoring rubrics, verified before they ever reach a user. I call it Gated Truth Architecture.

The Approach

Three expert AI personas, each with a defined scope: Neuro (nutrition science), Oracle (food history and culture), and Chefy (culinary technique). Every response is generated through a fine-tuned GPT-4.1-mini model trained on 10,000+ hand-curated seeds, knowledge units with strict schema, verified citations, and scoring rubrics.

Users ask a question. The system routes it to the right persona, grounds the answer in verified seeds, and delivers a response with traceable sources. No hallucinations. No hedging. No disclaimers that shift the burden back to the user.

The Result

Foodbe launched on the App Store in April 2026 with a $4.99/month subscription and 7-day free trial. 10,000+ hand-curated seeds, 500+ ingredient profiles, 190+ recipes, a live Developer API with 8 endpoints, and three published cookbooks.

Foodbe app screen 1
Foodbe app screen 2
Foodbe app screen 3
Foodbe app screen 4
Foodbe app screen 5
Foodbe app screen 6
Foodbe app screen 7

What I designed

  • Cross-platform design system — iOS, desktop web, mobile web with full feature parity
  • Conversational chat with persona switching, typing indicators, traceable citations
  • Spark chip system — 29 curated prompts that show users what Foodbe can do
  • Grocery list with point-of-decision insights per item
  • Food comparison — user defines what to compare
  • Meal plan with per-meal cost integration and grocery list generation
  • Recipe surface with diet tags, ingredient intelligence, persona tips inline
  • Onboarding — typewriter intro, persona introduction, spark chip discovery
  • Subscription flow — RevenueCat, 7-day trial, $4.99/month

What I built

  • Gated Truth Architecture — every response from a verified fact library
  • AI food intelligence engine with 3 expert personas — Neuro, Oracle, Chefy
  • Fine-tuned GPT-4.1-mini on 10,000+ hand-curated seeds
  • Brain API — 8 endpoints for external platforms
  • Developer portal — sandbox keys, API docs, real-time usage dashboard
  • Token system — bundles, pricing tiers, usage metering, deduction per API call
  • Failure tracking — every unknown routes to a new seed
  • Privacy controls — user data stored on device only, full data export available, option to disable all data storage

10,000+

hand-curated seeds

500+

ingredient profiles

190+

recipes

8

Brain API endpoints

3

published cookbooks

Gated Truth Architecture

AI wasn't built to tell the truth. That's not its goal.

Its goal is to pattern-match, to guess the best response to your input. Nowhere in its design is it told to validate from quality sources before it responds. That's a foundational problem.

Even the best RAG system runs at 91% accuracy. That means 1 in 11 answers is wrong. At 500 questions a day, that's ~3,000 wrong answers per month, silently delivered with the same confidence as correct ones.

GTA is different. It gates the AI and only lets it generate a response from a curated, verified fact library. The AI doesn't retrieve, it reasons from facts we know are true.

How the Foodbe brain works

01

Intent layer

Cleans input, determines intent — how, what, why

02

Expert routing

Routes to Neuro, Oracle, or Chefy

03

Seed matcher

Scores 10,000+ seeds using 4 methods — exact phrase, keyword, word, anchor

04

Fact library

Winning seed's verified anchors passed to fine-tuned model

05

Response

Model generates answer in persona voice with traceable citations

Gated Truth Architecture diagram — click to enlargeDownload GTA schema PDF ↓

A real seed — "Is oat milk healthy?"

{
  label: "Is oat milk healthy?",
  entity: "oat_milk_health",
  thesis: "Oat milk is legitimately healthy because its fiber
    actively lowers cholesterol and fortified versions match
    dairy's calcium. Commercial brands undermine this with
    rapeseed oil and added sugars.",

  neuro_anchors: [
    "beta-glucan forms viscous gels that bind bile acids
     in intestines",
    "liver pulls LDL cholesterol from blood to replace
     bound bile acids",
    "drops LDL ~5-7% at doses of 3g per day"
  ],

  neuro_citations: [
    { text: "beta-glucan LDL reduction ~6.5% meta-analysis
      of 28 RCTs",
      url: "https://pubmed.ncbi.nlm.nih.gov/25411276/" }
  ],

  oracle_anchors: [
    "Oatly co-founded in Sweden in 1994 by Rickard and
     Björn Öste to address lactose intolerance",
    "U.S. dairy industry received over $1.3B in federal
     payments in 2023 while plant milks receive none"
  ],

  chefy_anchors: [
    "froths well for coffee without curdling",
    "works best in recipes where slight sweetness fits"
  ],

  toneGates: { neuro: true, oracle: true, chefy: true }
}

Tools & how I work

I use many tools to design and build products.

Mainly I use my experience and knowledge. I also pair-program with Claude and ChatGPT.

Frontend

  • React Native (iOS)
  • Next.js
  • TypeScript
  • Tailwind CSS
  • Figma
  • Cursor
  • Vercel v0

Backend & infra

  • Node.js / Express
  • MongoDB
  • Railway
  • Vercel
  • AWS S3
  • Xcode

AI & model

  • OpenAI API
  • Fine-tuned GPT-4.1-mini
  • Gated Truth Architecture
  • Claude / Anthropic API
  • Prompt engineering
  • Agentic routing

Payments & subscriptions

  • Stripe
  • RevenueCat
  • Token bundles + usage metering

Analytics & auth

  • Mixpanel
  • Google Analytics
  • Firebase Auth

UX/UI

Prior work

2021–2024

GAINSystems

Founding Product Designer

Company's first dedicated product hire. Owned design across an enterprise SaaS suite used by global manufacturers, distributors, and retailers to forecast demand, optimize inventory, and manage supply chain performance.

  • Led a team of three developers across UX research, rapid prototyping, and UX/UI design, created the company's first design system and drove a complete platform redesign
  • Established a monthly user feedback panel with 10+ enterprise organizations to surface workflow breakdowns and drive product decisions before engineering handoff
  • Designed AI/ML product surfaces including Lead Time Prediction and the GAINS Assistant. During my tenure the company reported 47% ARR growth, 115% net retention, and 25+ new enterprise clients
Case Study

2015–2021

ArrowStream

Founding Product Designer

First dedicated product hire at North America's largest foodservice supply chain platform serving 100K+ operator locations, 1,500+ distributors, 260+ brands.

  • Created the company's first design system to drive product evolution across 130K+ restaurant locations
  • Architected and designed the Supplier Discovery marketplace, connecting buyers with 8,000+ suppliers and saving more than $5 million in buyer costs in its first year
  • Embedded with a team of 7 developers and wrote HTML and CSS for pixel-accurate, production-ready interfaces
Case Study

2011–2015

amSTATZ

Founding Product Designer

First dedicated product hire at an early-stage startup building health and fitness business management software for gyms and personal trainers.

  • Built the company's first design system and owned UI/UX and frontend development from day one
  • Conducted user research directly with personal trainers and gym owners, translating findings into user scenarios, use cases, wireframes, and high-fidelity prototypes
  • Embedded with the development team and wrote HTML5, CSS3, jQuery, and Bootstrap to ensure design intent was carried through to production
Case Study

Get in touch

You found the right person.

I design and build. I've been the founding designer at three software companies, always embedded with developers. Available now for founding designer, senior product designer, and AI product roles.

Based in Chicago. Open to local or remote work.