PlateProfitAI

PlateProfit AI v1.0.0-rc2 - Final Build Week Release Candidate

PlateProfitAI is a local-first Windows desktop application for restaurant costing, menu profitability, and evidence-backed decision support. It combines recipe costs, actual menu prices, weekly sales, menu structure, and ingredient price history so chefs and owners can see where profit is protected, where it is leaking, and what to review next.

The application runs offline, requires no account, and does not upload restaurant data.

Key Features

• Executive Dashboard with Business Health, weekly revenue, gross profit, risks, opportunities, and daily priorities
• Recipe Builder with unit conversion, yield-aware costing, Cost/Serve, Food Cost %, Profit/Serve, and suggested pricing
• Ingredient Directory, price updates, price history, trends, CSV import/export, and affected-dish analysis
• Dish Reports with actual and suggested pricing, cost drivers, pricing bands, and ingredient alternatives
• Menu Engineering matrix with Top Performers, Volume Drivers, Growth Opportunities, Review Candidates, confidence, and ranked actions
• Menu Portfolio category analysis and saved menu management
• Profit Simulator for read-only price, ingredient-cost, and weekly-sales scenarios
• Price Intelligence linking recorded ingredient movements to affected dishes
• Optimise Menu action queue for pricing, margin, recipe, and ingredient concentration reviews
• AI Menu Consultant with deterministic, explainable recommendations
• Text reports, CSV exports, local backup, and restore

Explainable Decision Engine

PlateProfitAI uses an explainable decision engine, not an external cloud LLM. Recommendations combine recipe cost, pricing, weekly sales, menu engineering, and recorded ingredient-price evidence. Financial values come from stored data and visible assumptions; unavailable results remain explicitly unquantified.

The engine does not claim customer-demand forecasting, POS integration, live supplier feeds, or autonomous pricing decisions.

Harbour & Hearth Bistro Showcase

The included deterministic Harbour & Hearth Bistro Showcase v1.0 contains:

• 68 ingredients
• 28 dishes across 8 categories
• 192 recipe items
• 138 ingredient price-history records
• 100% Actual Price, Weekly Sales, recipe-cost, and category coverage
• All four Menu Engineering quadrants

Use Load Demo Data on an empty installation. Existing user-owned records are preserved; repeated loading does not create duplicates. Clear Demo Data removes only demo-owned records. Reset is explicit and restores the curated baseline.

See BUILD_WEEK_SHOWCASE_DATA_GUIDE.md and BUILD_WEEK_DEMO_FLOW.md.

Packaged Quick Start

Download the RC2 judge package from the GitHub release:

PlateProfitAI v1.0.0-rc2 Build Week Release

1. Extract PlateProfitAI-v1.0.0-rc2-BuildWeek.zip.
2. Open PlateProfitAI.exe.
3. If Windows SmartScreen appears, choose More info and Run anyway. This unsigned release candidate is not code-signed.
4. Select Load Demo Data to explore the showcase.

Python is not required for the packaged application. See QUICK_START_BUILD_WEEK.md.

Run From Source

Validated with Python 3.14.4 on Windows.

cd "<path-to-PlateProfitAI>"
py -m venv .venv
.\.venv\Scripts\python.exe -m pip install -r requirements-release.txt
$env:PYTHONPATH = (Get-Location).Path
.\.venv\Scripts\python.exe -m plate_profit_ai.ui.desktop_app

Release Candidate Screenshots

Executive Dashboard	Menu Engineering

Profit Simulator	Optimise Menu

These images use the deterministic Harbour & Hearth Bistro Showcase dataset and contain no private restaurant data.

Data Locations

• Source-development mode: <project>\data
• Packaged Windows mode: %LOCALAPPDATA%\PlateProfitAI\data
• Optional test override: PLATEPROFIT_DATA_DIR
• Exports and backups: the location selected in the Windows save dialog

The packaged application does not store data beside the executable. Backup files are created only at a user-selected destination. Loading showcase data does not overwrite same-named user-owned records.

Judge Flow

1. Read Business Health and the next action on Executive Dashboard.
2. Open Menu Engineering and inspect all four quadrants.
3. In Profit Simulator, model a +$0.75 Double Bacon Burger price scenario.
4. Open Price Intelligence and inspect Chicken Thigh history and affected dishes.
5. Open a Dish Report or export the active analysis.

Simulator scenarios are session-only and read-only. They never silently update live dishes or ingredients.

How Codex and GPT-5.6 Were Used

PlateProfitAI was developed with Codex and GPT-5.6 as part of the Build Week workflow.

Codex

Codex was used directly throughout the engineering process to:

• Inspect and understand the existing repository
• Implement new application features and user-interface pages
• Refactor large modules into more maintainable components
• Trace calculation and terminology inconsistencies across the application
• Create regression tests for costing, profitability, menu analysis and storage
• Run compile checks, automated tests and desktop import smoke tests
• Build the standalone Windows executable with PyInstaller
• Audit the final source and judge packages for private or unintended files
• Synchronise the final RC2 source and release documentation with GitHub

GPT-5.6

GPT-5.6 was used for product reasoning and development planning, including:

• Identifying the restaurant problems PlateProfitAI should prioritise
• Designing workflows for chefs, restaurant owners and non-technical users
• Reviewing recommendation logic, risk, confidence and financial terminology
• Improving Menu Engineering labels and explanations
• Planning the Executive Dashboard and Profit Simulator demonstrations
• Structuring Codex implementation prompts and validation tasks
• Reviewing the final demo flow, narration and Devpost submission materials

The runtime application does not require Codex, GPT-5.6 or an external language model. PlateProfitAI’s AI Menu Consultant is a deterministic, local and explainable decision engine based on saved recipe, pricing, sales, category and ingredient-price evidence.

Testing

Latest RC2 validation: 625 passing tests.

py -m compileall plate_profit_ai tests scripts
py -m pytest -q
py -c "from plate_profit_ai.ui.desktop_app import PlateProfitDesktopApp; print('Desktop app import passed')"

Reproducible Windows Build

Pinned dependencies are in requirements-release.txt.

powershell -ExecutionPolicy Bypass -File scripts\build_release_candidate.ps1

The script creates a clean release environment, validates source, builds with PyInstaller, packages judge/source archives, generates checksums, and audits contents.

Technology

• Python
• CustomTkinter / Tkinter
• Matplotlib for contained analytical visuals
• Pillow for branded image assets
• PyInstaller for the standalone Windows executable
• JSON and CSV local storage

Known Limitations

• Windows release candidate; unsigned executable may trigger SmartScreen
• Manual data entry; no POS, invoice scanning, supplier API, cloud sync, or multi-location support
• Profit Simulator scenarios are session-only and do not model demand elasticity
• Recommendations are deterministic estimates and should be reviewed by the chef or owner
• Optimise Menu's review score measures optimisation pressure, not overall Business Health

Licence And Attribution

See LICENSES_OR_ATTRIBUTIONS.md. No separate project-source licence has been declared in this release candidate.