Gordon

Inspiration: From "Fridge Fatigue" to Culinary Confidence

For many young adults and busy professionals, the kitchen is no longer a place of creativity—it’s a source of decision fatigue. We observed a recurring cycle: the "Sunday Meal Prep" that leads to eating the same soggy pasta for five days, the "Ingredient Amnesia" where fresh greens turn into science experiments in the crisper drawer, and the inevitable "Takeout Defeat" when planning feels like a second job.

Gordon was born to bridge the gap between having ingredients and having a plan. We didn't just want a recipe app; we wanted an intelligent agent that understands the state of your kitchen as well as your personal cravings.

What Gordon Does: Your End-to-End Kitchen OS

Gordon is an intelligent cooking ecosystem that solves the "What’s for dinner?" problem before you even ask it.

The Smart Fridge: Gordon tracks your inventory and uses predictive logic to surface ingredients nearing their expiration.

Intelligent Discovery: Unlike static recipe sites, Gordon’s recommendations are a three-way intersection of what you have, what you need to use, and what you actually want to eat.

Automated Logistics: Once a meal is chosen, Gordon’s "sous chefs" handle the dirty work—generating shopping lists, checking price estimates, and integrating with Instacart/Amazon Fresh for one-click fulfillment.

How we built it

We built Gordon using a Modular Agent Orchestration pattern on the Fetch.ai ASI:One framework. Rather than a single, bloated LLM prompt, we utilized In-Context Learning (ICL) to create a specialized workforce:

The Workforce Breakdown:

Gordon (The Orchestrator): High-creativity (temp 0.8) agent focused on recipe ideation and user interaction.

The Curator (Inventory Agent): Precision-focused (temp 0.3) agent that performs "inventory diffing" to see what’s missing.

The Visionary (OCR Agent): Powered by asi1-mini, this agent extracts structured data from grocery receipts to update inventory instantly.

The Logistics Agent: A deterministic-generative hybrid that builds shopping URLs and manages the MongoDB Atlas state.

Challenges we ran into

Our journey revealed that "reasoning about food" is one of the most nuanced challenges for AI. We learned that:

Food Quality is Heuristic: Since an AI can't smell the milk, we developed "Freshness Heuristics"—predictive models based on storage conditions and typical shelf life.

What we are proud of

Gordon serves as a prime example of the Fetch.ai vision: a world where autonomous agents collaborate to solve daily human problems.

Practical ICL Implementation: We demonstrated that ASI:One can handle multi-modal tasks (Vision + Reasoning) in a single pipeline.

Real-World Utility: Gordon tackles the $400+ billion problem of annual food waste, proving that agentic workflows have massive environmental and economic ROI.

What we learned

Prompt Engineering is the New Training: We proved that you don't need custom-trained models to build sophisticated tools. By using ASI:One’s ICL capabilities, we achieved high-level reasoning through strategic context-passing and temperature tuning.

Friction is the Enemy: The hardest part of food management is data entry. This led us to focus on Receipt OCR and Progressive Disclosure to keep users engaged without making them feel like they were doing data entry.

What's next for Gordon

Seamless Inventory Intelligence

The primary goal is to lower the barrier for inventory tracking, moving from manual entry to passive collection.

Next-Gen Input Methods: Beyond simple receipt OCR, we are exploring Real-Time Computer Vision for pantry scanning and integration with Smart Fridge APIs (like Samsung’s Bespoke AI 2026 series) to automatically sync "fridge views" with ingredient lists.

Direct Retailer Hooks: Developing deep integrations with retail loyalty programs to pull digital receipts directly into Gordon’s database the moment a user checks out.

Personalized "Brain" & Recommendation Logic

We are shifting from static recipe matching to a dynamic, behavioral recommendation engine.

Taste Profile Learning: Algorithms that learn not just what you can cook, but what you love—prioritizing regional flavors, favorite cuisines, and even "lazy day" vs. "ambitious weekend" cooking patterns.

Hyper-Local Awareness: Integrating seasonal ingredient data and local grocery store flyers to recommend recipes based on what is currently freshest and most cost-effective in the user's specific zip code.

Deep Commerce Integration

The future of Gordon is moving away from simple search queries toward direct, programmatic fulfillment.

Native Cart Management: Utilizing the latest Amazon Business Cart APIs and Instacart Connect to allow users to modify, price-compare, and finalize grocery orders without ever leaving the Gordon interface.

Automated "Stock-Up" Logic: Implementing agents that can predict when "staple" ingredients (milk, oil, spices) are low and suggest adding them to the weekly meal-prep order before the user even notices they're gone.

Built With

• asi:one
• fastapi
• fetch.ai
• mongodb
• react
• tailwindcss
• typescript