AI Project Concept Template for Windesheim/VCH

This template is for planning AI project ideas before fully committing to them. Use this to structure your thinking, identify requirements, and decide if the project is worth pursuing.

All projects follow Agile principles with iterative development, regular sprint cycles, and continuous stakeholder feedback.

Instructions

1. Copy this template for each new AI project concept
2. Fill in as much as you know - it’s OK to leave sections blank or mark as “TBD”
3. Save with a descriptive filename (e.g., ai-chocolate-traceability-concept.md)
4. Create a matching folder: assets/projects/[project-name]/ for images and visual assets
5. Projects can be added to the website with “Concept” status to gauge interest
6. Once approved and resourced, convert to a full project

VISUAL ASSETS & BRANDING

Project Folder: assets/projects/[project-name]/

Required Assets:

• hero-image.jpg or hero-image.png (1200x600px recommended) - Main visual for website
• logo.png or icon.svg (if project has its own branding)
• thumbnail.jpg (400x300px) - For project cards on homepage

Optional Assets:

• diagram-architecture.png - System architecture or data flow
• screenshot-*.png - Screenshots of prototypes, dashboards, interfaces
• mockup-*.png - Design mockups or wireframes
• presentation-deck.pdf - Pitch deck or project overview slides
• demo-video-thumbnail.jpg - If you have a demo video

Asset Checklist:

assets/projects/[project-name]/
├── hero-image.png          ← Main banner image
├── thumbnail.png           ← Small card preview
├── logo.svg               ← Project logo (optional)
├── diagrams/
│   └── architecture.png
├── screenshots/
│   ├── dashboard.png
│   └── mobile-app.png
└── README.md              ← Asset documentation

PROJECT CONCEPT

Basic Information

Project Title:

Project Slug:

Concept Status:

Category:

THE 3 MINUTE RULE PITCH

(Answer these 6 questions from Brant Pinvidic’s framework)

1. What is it?

2. How does it work?

3. Are you sure?

4. Can you do it?

5. Can you make money? (or: What’s the value?)

6. Are there any risks?

The Problem

What problem does this solve?

Who has this problem?

How big is the problem?

The AI Solution

What AI/ML technique would you use?

Why does this need AI?

What data would you need?

• Data type 1:
• Data type 2:
• Data availability: (Do we have it? Can we get it? How hard?)

Expected output/deliverable:

Feasibility Check

Technical Difficulty:

Required Skills:

Resources Needed:

• Computing:
• Data:
• People:
• Budget:

TIMELINE & MILESTONES (Agile Planning)

Project Duration

Total Estimated Time:

Sprint Length:

Total Sprints:

High-Level Timeline

Phase 1: Discovery & Setup (Estimated: ___ weeks)

• Sprint 1-2: Project kickoff, requirements gathering, data exploration
• Sprint 3: Environment setup, initial data pipeline

Phase 2: MVP Development (Estimated: ___ weeks)

• Sprint 4-6: Core model development
• Sprint 7: First working prototype

Phase 3: Iteration & Enhancement (Estimated: ___ weeks)

• Sprint 8-10: Feature additions, model improvements
• Sprint 11: User testing and feedback integration

Phase 4: Delivery & Handoff (Estimated: ___ weeks)

• Sprint 12: Documentation, final testing, deployment

Detailed Milestone Plan

Customize this table for your specific project. Add or remove sprints as needed.

Agile Ceremonies

Daily Standups:

Sprint Planning:

Sprint Review/Demo:

Sprint Retrospective:

Backlog Grooming:

Key Dates & Deadlines

• Project Start Date:
• First Demo: (Sprint 7 end)
• User Testing Begins: (Sprint 10)
• Final Presentation:
• Project End Date:
• External Deadlines:

Strategic Fit

How does this align with VCH goals?

Potential Partners:

Student Learning Value:

Reusability:

Success Criteria

How would you know if this worked?

Minimum Viable Product (MVP):

Stretch Goals:

Risks & Unknowns

What could go wrong?

What don’t you know yet?

Deal-breakers:

Next Steps

What needs to happen before this becomes a real project?

Who needs to approve this?

Timeline for decision:

EXAMPLE FILLED OUT

Here’s a complete example:

VISUAL ASSETS & BRANDING

Project Folder: assets/projects/ai-coffee-quality/

Required Assets:

• hero-image.png - Coffee farm landscape with data overlay
• thumbnail.png - Coffee beans and graph icon
• logo.svg - Not needed yet

Optional Assets:

• diagrams/model-architecture.png - Shows data flow from farm to prediction
• screenshots/dashboard-mockup.png - Wireframe of farmer-facing app

PROJECT CONCEPT

Basic Information

Project Title: AI-Powered Coffee Quality Prediction from Farm Data

Project Slug: ai-coffee-quality

Concept Status: under-review

Category: student-project, partner-collaboration

THE 3 MINUTE RULE PITCH

1. What is it?

A mobile tool that helps coffee farmers predict their harvest quality before it happens, using machine learning trained on past farm and processing data.

2. How does it work?

Farmers input details about their farm (altitude, rainfall) and how they process the coffee (fermentation time, drying method). The AI model compares this to thousands of past harvests and predicts what quality score they’ll likely get. Then it suggests changes they could make to improve.

3. Are you sure?

Similar models exist for wine grape quality prediction with 85%+ accuracy. Coffee cooperatives already track this data, we just need to partner with one. Research papers show farm conditions strongly correlate with cup quality.

4. Can you do it?

We have two computer science students who completed the ML course. Christiaan has supply chain connections to get coffee data. We can start with a simple model in Python and test accuracy before building the app.

5. What’s the value?

For farmers: better prices if they can improve quality, less income uncertainty. For students: real ML project with actual users and measurable impact. For coffee importers: better supply chain transparency and consistency.

6. Are there any risks?

• Might not get enough historical data (need 200+ samples)
• Model accuracy might be too low to be useful
• Farmers might not trust AI predictions over their experience

The Problem

What problem does this solve? Coffee farmers don’t know their quality score until after harvest, roasting, and cupping (tasting). By then it’s too late to fix problems. This causes income uncertainty and makes it hard to improve.

Who has this problem? Smallholder coffee farmers in Ethiopia, Kenya, Colombia who sell specialty coffee. Also affects coffee importers who buy based on quality.

How big is the problem? Affects 100,000+ smallholder specialty coffee farmers globally. Quality differences can mean 2x price difference.

The AI Solution

What AI/ML technique would you use? Supervised learning (Random Forest or XGBoost) to predict cupping scores from farm and processing data.

Why does this need AI? Complex interactions between 20+ variables (altitude, rainfall, processing method, fermentation time, drying method). Traditional formulas don’t work well.

What data would you need?

• Historical cupping scores (quality ratings) from past harvests
• Farm data: altitude, rainfall, soil type, shade coverage
• Processing data: fermentation hours, drying method, storage time
• Data availability: Coffee cooperatives and importers track this. Need partnership to access.

Expected output/deliverable: Mobile app where farmers input their farm and processing choices, get predicted quality score and suggestions for improvement.

Feasibility Check

Technical Difficulty: Medium

Required Skills:

• Python, scikit-learn basics
• Data cleaning and feature engineering
• Some supply chain knowledge (can be learned)
• Mobile app development (or start with web app)

Resources Needed:

• Computing: Standard laptop is fine
• Data: Need partnership with coffee importer or cooperative
• People: 2 students (one for ML, one for app) + supervision
• Budget: €0 (all open-source tools)

TIMELINE & MILESTONES (Agile Planning)

Project Duration

Total Estimated Time: 6 months (September 2025 - February 2026)

Sprint Length: 2 weeks

Total Sprints: 12 sprints

High-Level Timeline

Phase 1: Discovery & Setup (6 weeks)

• Sprint 1-2: Partner meetings, data access negotiation, requirements
• Sprint 3: Python environment, initial data exploration

Phase 2: MVP Development (8 weeks)

• Sprint 4-6: Model development, testing different algorithms
• Sprint 7: First working model that beats baseline

Phase 3: Iteration & Enhancement (6 weeks)

• Sprint 8-9: Model improvements, feature engineering
• Sprint 10: Build simple web interface for testing

Phase 4: Delivery & Handoff (4 weeks)

• Sprint 11: User testing with 5 farmers, collect feedback
• Sprint 12: Final documentation, code cleanup, handoff

Detailed Milestone Plan

Agile Ceremonies

Daily Standups: Every Monday/Wednesday/Friday, 15 min async on Discord

Sprint Planning: First Monday of each sprint, 1 hour on Zoom

Sprint Review/Demo: Last Friday of each sprint, show progress to Christiaan and partner

Sprint Retrospective: After sprint review, 30 min team-only reflection

Backlog Grooming: Mid-sprint, review and prioritize upcoming tasks

Key Dates & Deadlines

• Project Start Date: September 1, 2025
• First Demo: November 24 (Sprint 7 end)
• User Testing Begins: February 2 (Sprint 11)
• Final Presentation: February 28, 2026
• Project End Date: February 28, 2026
• External Deadlines: None (internal project)

Strategic Fit

How does this align with VCH goals?

• Practical AI application for agricultural supply chains
• Helps smallholder farmers (social impact)
• Good student learning project (real data, real impact)
• Could become reusable tool for other crops

Potential Partners:

• ClearCaff (coffee data platform)
• Local coffee roasters with direct trade relationships
• Coffee cooperatives in Ethiopia

Student Learning Value: High - covers data cleaning, feature engineering, model selection, deployment, user research with farmers.

Reusability: Medium - could adapt for cacao, tea, wine grapes with similar approach.

Success Criteria

How would you know if this worked?

• Model predicts cupping scores within ±3 points (on 100-point scale)
• At least 5 farmers use the app and find it helpful
• Model identifies at least 2 actionable factors farmers can control

Minimum Viable Product (MVP): Python script that takes CSV input and outputs prediction. No app needed initially.

Stretch Goals:

• Full mobile app with translations
• Regional models (Ethiopia vs Colombia might need different models)
• Integration with farm management software
• Recommendations engine (“try fermenting 2 hours longer”)

Risks & Unknowns

What could go wrong?

• Can’t find partner willing to share cupping score data (it’s sensitive business info)
• Not enough data samples (need 200+ for decent model)
• Variables we can collect don’t actually predict quality well
• Farmers don’t trust or use the predictions

What don’t you know yet?

• What data is actually available from partners
• How cupping scores are measured consistently
• Whether farmers would actually change practices based on predictions

Deal-breakers:

• Can’t get data with at least 200 samples
• Model accuracy is worse than random guessing
• No farmers interested in using it

Next Steps

What needs to happen before this becomes a real project?

1. Talk to 2-3 coffee importers about data availability
2. Literature review - has anyone done this already?
3. Find 2 interested students
4. Confirm partner commitment

Who needs to approve this? Christiaan + potential coffee partner

Timeline for decision: 2 months - decide by end of Q1 2025

NOTES

Tips for good AI project concepts:

• Start with the problem, not the AI technique
• Be honest about what you don’t know
• Smaller and focused is better than big and vague
• Consider if this really needs AI or if simpler solutions exist
• Think about who would actually use this

Common mistakes to avoid:

• “AI can solve everything” thinking
• Underestimating data requirements
• Not checking if someone already built this
• Ignoring the deployment/usage challenge
• Choosing trendy tech that doesn’t fit the problem

Agile Best Practices:

• Keep sprints consistent (usually 2 weeks)
• Don’t skip retrospectives - they’re where you improve
• Demo working software every sprint, even if it’s rough
• Adapt the plan as you learn - timelines will change
• Focus on delivering value, not checking boxes
• Make blockers visible early

Managing Visual Assets:

• Create the project folder structure early, even if you don’t have all assets yet
• Use consistent naming: hero-image.png, thumbnail.png, screenshot-dashboard.png
• Keep original high-res files in a separate folder
• Document what each image shows in the folder README
• Update assets as the project evolves (mockup → screenshot → final)
• Use placeholder images if needed to visualize the website early