e
Can AI tell us when a cacao story should not be trusted?
Inspiration
Cacao has a story.
Where was it grown? Who grew it? When was it harvested? How was it grown? What environmental conditions affected it?
Today, much of this information is fragmented across farmers, buyers, spreadsheets, photos, messages, and paper records.
The problem is not only lack of information.
The bigger problem is:
How do we know which claims can actually be trusted?
Open Cacao explores a different approach to agricultural traceability: instead of simply recording a farmer's story, an AI agent investigates the story, identifies supporting evidence, detects missing information, and communicates its confidence.
What is Open Cacao?
Open Cacao is an AI-powered trust layer for cacao.
A farmer, cooperative, or buyer can provide information about a cacao farm or harvest.
The Open Cacao Agent then:
- Understands the cacao claim
- Breaks the claim into verifiable information
- Collects available evidence
- Checks consistency between claims and evidence
- Identifies missing or conflicting information
- Produces a confidence assessment
- Generates a structured Cacao Trust Passport
Instead of simply saying:
"This is organic cacao."
the system can say:
Organic claim: UNVERIFIED Evidence found: Farmer statement Independent evidence: Missing Confidence: Low
The goal is not for AI to blindly approve a story.
The goal is for AI to make the evidence behind the story visible.
The Problem
Agricultural traceability systems often focus on recording events:
Plant → Grow → Harvest → Sell
But a record is not automatically proof.
For example:
"This farm is organic."
is a claim.
A photo is evidence.
A certification document is evidence.
A location is evidence.
A climate record is evidence.
A consistent timeline is evidence.
Open Cacao asks:
Can an AI agent connect these pieces of evidence and explain what is known, what is uncertain, and what is missing?
How It Works
1. Start with a cacao farm
The user provides basic information such as:
- Farm location
- Farmer information
- Crop information
- Planting/harvest information
- Farming claims
- Photos or other evidence
2. The AI Agent investigates
The agent identifies the claims that require verification.
For example:
Claim: Organic cacao
The agent asks:
- What evidence supports this?
- Is there certification?
- Is there a documented farming history?
- Is the location consistent?
- Are there contradictory records?
- What evidence is still missing?
3. Evidence Layer
Every important claim receives an evidence state:
Verified
Evidence sufficiently supports the claim.
Partially supported
Some evidence exists, but important information is missing.
Unverified
The claim currently relies mainly on an unsupported statement.
Conflicting
Available evidence does not agree.
4. Cacao Trust Passport
The final output is a structured passport showing:
- Farm identity
- Cacao origin
- Crop information
- Evidence timeline
- Claims
- Evidence supporting each claim
- Missing evidence
- Confidence level
- AI-generated explanation
The passport can be shared with a buyer.
What Makes It Different?
Existing digital crop passport concepts focus on documenting the journey of a crop.
Open Cacao focuses on a different question:
Can we evaluate the trustworthiness of the story being documented?
We are moving from:
Traceability
to:
Evidence intelligence
and ultimately:
Agricultural trust.
Why Cacao?
Cacao is our first deep-domain use case because its supply chain contains a difficult combination of:
- Smallholder farmers
- Fragmented farm data
- Sustainability claims
- Geographic origin
- Climate exposure
- Quality information
- Multiple actors between farmer and buyer
Rather than building a generic agricultural chatbot, Open Cacao deliberately goes deep into one crop and one supply-chain problem.
Why AI Agents?
A conventional application waits for the user to enter information.
Open Cacao is designed around an agentic workflow.
The agent can:
Understand → investigate → compare → identify gaps → assess → explain
This allows the AI to become an investigator rather than simply a chatbot.
Example
A farmer submits:
"Organic cacao, harvested June 2026."
Open Cacao investigates the claim.
Result
Organic
🟡 Unverified
Harvest date
🟢 Supported by farm record
Farm location
🟢 Supported
Independent certification
🔴 Missing
Evidence confidence
64%
AI explanation
The available farm records support the harvest information and farm location. The organic claim currently depends on the farmer's statement and does not contain independent certification evidence.
The system does not invent certainty.
It shows uncertainty.
Built With
- OpenAI Codex
- AI agent workflow
- Web application
- Structured agricultural data
- Evidence evaluation
- Cacao domain knowledge
Codex is used throughout the development process to accelerate implementation, testing, debugging, iteration, and integration.
Why Codex?
Open Cacao is a Deep Domain AI problem.
The challenge is not simply generating code.
The challenge is translating a complex real-world domain into an AI workflow that can:
- reason about claims
- work with structured and unstructured information
- identify evidence
- handle uncertainty
- produce an understandable result
Codex allows the team to rapidly prototype and iterate this workflow during the hackathon.
What We Built During the Hackathon
The hackathon version focuses on one critical workflow:
Give Open Cacao a cacao farm and a set of claims. Let the AI investigate the available evidence and produce a Trust Passport.
We deliberately avoid building a large agricultural management system.
One workflow.
One domain.
One difficult problem.
Trust.
Vision
Open Cacao starts with cacao.
But the underlying problem is much larger.
Farmers around the world make claims about:
- origin
- sustainability
- farming practices
- quality
- climate resilience
Buyers need to understand which claims are supported and which remain uncertain.
Our long-term vision is:
An AI trust layer for agricultural supply chains.
Cacao is where we start.
Final Thought
AI should not simply make agricultural stories easier to tell.
It should make them harder to fake and easier to understand.
Open Cacao — From crop stories to trusted evidence.
Built With
- dex
Log in or sign up for Devpost to join the conversation.