Branding
From leftovers to recipes in seconds
How a conversational AI assistant turns leftovers into meals and conversations into learning.
Year
2024-2025
Industry
Food & Beverage
Timeframe
5 months
My Role
UX Conversation Designer, AI Systems Designer, Prompt Engineer
Problem
The client wanted to innovate and explore how AI could improve their users’ experience with a large recipe archive. Their platform to connect with users is a website filled with thousands of recipes, but they wanted something more conversational, a way for people to simply say “I have these leftovers” or “What can I cook with this?” and receive tailored suggestions.
The goal was to make seasonality and food-waste awareness part of a smoother, friendlier user journey.
Solution
The project began with a stakeholder workshop to align on needs and scope. The team defined a proof-of-concept chatbot that:
Treated every user query as recipe-related, ensuring the assistant always guided conversations toward meal inspiration.
Delivered personalized, engaging suggestions with a focus on seasonal ingredients and user-specified criteria (e.g., cooking time, dietary needs).
Prioritized a verified recipe knowledge base as the single source of truth, linking every recipe directly and transparently.
Prevented hallucinations by serving exact ingredients and preparation steps only from validated entries, and clearly flagged AI-generated recipes when no match existed.
Designed conversational continuity with follow-up prompts, context retention, and natural dialogue to keep users engaged.
The MVP was built in under three months using GPT-4o Mini, retrieval-augmented generation with Postgres/pgvector, and an internal AI solution framework. It was soft-launched on a dedicated subdomain, activated via an avatar teaser, and instrumented with analytics and transcript review for iterative improvement.
Challenge
The main challenges were ensuring brand consistency in every conversation, implementing guardrails to block unwanted topics and safeguard against extreme misuse, preventing hallucinated recipes through a strict card-and-link pattern, and aligning all transcript storage and analytics with GDPR compliance requirements.
Summary
The project transformed a broad ambition into a working conversational AI assistant that connects leftover ingredients with trusted recipes. By focusing on food waste and seasonality, it introduced a more natural way for users to engage with the recipe archive. Beyond improving the cooking experience, the chatbot also provides valuable insight into users’ needs and behaviors through their conversations, showing how AI can both enhance everyday life and generate learning for continuous improvement.
More Projects
Branding
From leftovers to recipes in seconds
How a conversational AI assistant turns leftovers into meals and conversations into learning.
Year
2024-2025
Industry
Food & Beverage
Timeframe
5 months
My Role
UX Conversation Designer, AI Systems Designer, Prompt Engineer
Problem
The client wanted to innovate and explore how AI could improve their users’ experience with a large recipe archive. Their platform to connect with users is a website filled with thousands of recipes, but they wanted something more conversational, a way for people to simply say “I have these leftovers” or “What can I cook with this?” and receive tailored suggestions.
The goal was to make seasonality and food-waste awareness part of a smoother, friendlier user journey.
Solution
The project began with a stakeholder workshop to align on needs and scope. The team defined a proof-of-concept chatbot that:
Treated every user query as recipe-related, ensuring the assistant always guided conversations toward meal inspiration.
Delivered personalized, engaging suggestions with a focus on seasonal ingredients and user-specified criteria (e.g., cooking time, dietary needs).
Prioritized a verified recipe knowledge base as the single source of truth, linking every recipe directly and transparently.
Prevented hallucinations by serving exact ingredients and preparation steps only from validated entries, and clearly flagged AI-generated recipes when no match existed.
Designed conversational continuity with follow-up prompts, context retention, and natural dialogue to keep users engaged.
The MVP was built in under three months using GPT-4o Mini, retrieval-augmented generation with Postgres/pgvector, and an internal AI solution framework. It was soft-launched on a dedicated subdomain, activated via an avatar teaser, and instrumented with analytics and transcript review for iterative improvement.
Challenge
The main challenges were ensuring brand consistency in every conversation, implementing guardrails to block unwanted topics and safeguard against extreme misuse, preventing hallucinated recipes through a strict card-and-link pattern, and aligning all transcript storage and analytics with GDPR compliance requirements.
Summary
The project transformed a broad ambition into a working conversational AI assistant that connects leftover ingredients with trusted recipes. By focusing on food waste and seasonality, it introduced a more natural way for users to engage with the recipe archive. Beyond improving the cooking experience, the chatbot also provides valuable insight into users’ needs and behaviors through their conversations, showing how AI can both enhance everyday life and generate learning for continuous improvement.
More Projects
Branding
From leftovers to recipes in seconds
How a conversational AI assistant turns leftovers into meals and conversations into learning.
Year
2024-2025
Industry
Food & Beverage
Timeframe
5 months
My Role
UX Conversation Designer, AI Systems Designer, Prompt Engineer
Problem
The client wanted to innovate and explore how AI could improve their users’ experience with a large recipe archive. Their platform to connect with users is a website filled with thousands of recipes, but they wanted something more conversational, a way for people to simply say “I have these leftovers” or “What can I cook with this?” and receive tailored suggestions.
The goal was to make seasonality and food-waste awareness part of a smoother, friendlier user journey.
Solution
The project began with a stakeholder workshop to align on needs and scope. The team defined a proof-of-concept chatbot that:
Treated every user query as recipe-related, ensuring the assistant always guided conversations toward meal inspiration.
Delivered personalized, engaging suggestions with a focus on seasonal ingredients and user-specified criteria (e.g., cooking time, dietary needs).
Prioritized a verified recipe knowledge base as the single source of truth, linking every recipe directly and transparently.
Prevented hallucinations by serving exact ingredients and preparation steps only from validated entries, and clearly flagged AI-generated recipes when no match existed.
Designed conversational continuity with follow-up prompts, context retention, and natural dialogue to keep users engaged.
The MVP was built in under three months using GPT-4o Mini, retrieval-augmented generation with Postgres/pgvector, and an internal AI solution framework. It was soft-launched on a dedicated subdomain, activated via an avatar teaser, and instrumented with analytics and transcript review for iterative improvement.
Challenge
The main challenges were ensuring brand consistency in every conversation, implementing guardrails to block unwanted topics and safeguard against extreme misuse, preventing hallucinated recipes through a strict card-and-link pattern, and aligning all transcript storage and analytics with GDPR compliance requirements.
Summary
The project transformed a broad ambition into a working conversational AI assistant that connects leftover ingredients with trusted recipes. By focusing on food waste and seasonality, it introduced a more natural way for users to engage with the recipe archive. Beyond improving the cooking experience, the chatbot also provides valuable insight into users’ needs and behaviors through their conversations, showing how AI can both enhance everyday life and generate learning for continuous improvement.