PlateProfitAI started from a real conversation with a chef friend about how difficult it can be to understand whether a dish is genuinely profitable.
Supplier prices change, recipe costs are often scattered across spreadsheets, and a dish that sells well can still have weak margins. I wanted to build something that connected ingredient costs, menu prices, weekly sales and supplier movements in one place, then converted that information into practical decisions for chefs and restaurant owners.
What it does
PlateProfitAI is a local-first Windows application for restaurant costing and menu profitability intelligence.
It allows a restaurant to:
- Track ingredients, pack prices, yields and supplier-price history
- Calculate recipe cost and cost per serving
- Compare actual menu prices with suggested pricing
- Measure food-cost percentage and profit per serve
- View weekly revenue and gross-profit performance
- Identify supplier-price risks and affected dishes
- Analyse the menu using popularity and profit contribution
- Group dishes into Top Performers, Volume Drivers, Growth Opportunities and Review Candidates
- Rank quantified financial opportunities separately from strategic reviews
- Model proposed price or cost changes in a read-only Profit Simulator
- Review category performance through Menu Portfolio analysis
- Generate evidence-backed guidance through the AI Menu Consultant
- Export reports and create or restore local backups
The included showcase menu demonstrates the full workflow with dishes, recipes, selling prices, weekly sales, categories and ingredient-price history.
How I built it
PlateProfitAI was built in Python as a modular Windows desktop application.
The interface uses CustomTkinter, with Matplotlib used for charts and visual analysis. Data is stored locally using JSON, and the packaged Windows executable was created with PyInstaller.
The project separates:
- Data models and storage
- Recipe and profitability calculations
- Menu Engineering analysis
- Optimisation and opportunity ranking
- Price Intelligence
- Profit simulation
- Dashboard aggregation
- Desktop user-interface pages
- Reporting, export, backup and restore
A major focus was keeping every financial value consistent across the application. Profit, food cost, weekly contribution, scenario impact, risk and confidence each have separate definitions so that the same number is not given different meanings on different pages.
The current release passes 625 automated tests and includes a packaged Windows build that does not require Python to be installed.
How I used Codex and GPT-5.6
I used Codex throughout the main development workflow to inspect the repository, implement features, refactor modules, trace calculation inconsistencies, write regression tests, rebuild the executable and audit the final release packages.
GPT-5.6 helped me reason through product decisions such as:
- Which restaurant problems were most valuable to solve
- How recommendations should be ranked
- How risk and confidence should be communicated
- How to avoid double-counting overlapping opportunities
- How unknown financial outcomes should be displayed honestly
- How to simplify professional terminology for non-technical users
- How to structure the final judge demonstration
Codex was also used to run large validation passes, including compile checks, automated testing, GUI smoke tests, package inspection and release preparation.
PlateProfitAI does not depend on an external language model at runtime. Its AI Menu Consultant is an explainable decision engine that works from local recipe, price, sales, category and supplier-cost evidence. It does not invent missing financial values.
Challenges
One of the hardest challenges was maintaining consistency across a large number of connected modules.
A change to an ingredient price can affect recipe cost, dish profitability, supplier alerts, menu optimisation, category performance and the Executive Dashboard. Every page needed to use the same underlying definitions.
Other challenges included:
- Handling incomplete data without presenting fabricated results
- Distinguishing current profit from projected improvement
- Separating risk from confidence
- Preventing overlapping dish and category opportunities from being added twice
- Making detailed restaurant analysis understandable to non-technical users
- Packaging a reliable standalone Windows application
- Creating a realistic showcase menu that exercised every major workflow
Accomplishments that I am proud of
- Built a complete packaged Windows desktop application
- Reached 625 passing automated tests
- Created a full restaurant showcase with menu, recipe, sales and supplier-price data
- Built an Executive Dashboard that converts data into ranked daily actions
- Created a read-only Profit Simulator with weekly, monthly and annual estimates
- Built Menu Engineering, Menu Portfolio and Price Intelligence workflows
- Ensured recommendations show their supporting evidence
- Kept unknown outcomes labelled as unquantified rather than inventing dollar values
- Created clean judge and source release packages with automated package audits
- Produced a complete demo and testing workflow for the final release candidate
What I learned
I learned that useful business intelligence is not only about producing more calculations.
The results must be:
- Consistent
- Explainable
- Actionable
- Honest about missing information
- Easy for the intended user to understand
I also learned how important testing becomes when the same financial values appear across many different screens and reports. Small wording or calculation inconsistencies can reduce trust even when the underlying product works.
What is next
Future development could include:
- Supplier invoice importing
- POS and inventory integrations
- Persistent saved simulator scenarios
- Multi-location restaurant support
- Longer-term sales and profitability trends
- Team accounts and approval workflows
- Improved accessibility and onboarding
- Additional supplier and menu-data import formats
The current version focuses on providing a reliable, local-first foundation for practical restaurant profitability decisions.
Log in or sign up for Devpost to join the conversation.