Inspiration
I wanted a phone-friendly application for the keto world: reading carbohydrate information from products, finding restaurant options nearby, and finding products to buy. I chose the name Keto World and a professional design in Spanish from Spain.
What it does
Keto World processes a photographed nutrition label using local Tesseract.js OCR. It proposes a carbohydrate value per explicit 100 g or 100 ml and asks the visitor to check and confirm it against the photo. Ambiguous columns or missing units require review; sugars are never substituted for total carbohydrates. A reset removes the photo and extracted text.
The restaurant section searches a small catalogue of dated, reviewed official menus around Mijas and Fuengirola. Located entries can be filtered within ten kilometres; named-area queries and Google Maps links are also available. The app does not certify restaurants or invent nutritional totals. Browser dictation and spoken summaries have a typing fallback. Shopping opens Amazon.es or other-store searches; no affiliate earnings are active.
How it was built
I supplied the idea, priorities, name and style, and delegated implementation to an AI coding assistant. We used the Devpost Learn Skill Pack and scope.md, prd.md and spec.md to define a small proof of concept before implementation. New HTML, CSS and JavaScript code was created for this project. OCR code and language models are bundled with their third-party licences. Photos and OCR text stay in browser memory. No paid API, database, app account or payment system is required.
Challenges and validation
Live restaurant-map requests failed verification, so the dependency was removed and replaced with an explicit, local catalogue. OCR can misread small text, so the parser rejects ambiguous results and a person must verify the label. Thirteen automated tests passed. Browser checks covered real OCR on a clearly fictional sample label, correction/confirmation, reset, named-area lookup and shopping links. The demo is an edited walkthrough of real application screenshots, not a live phone recording.
Learning and current limits
This is an early experiment in turning my product requirements into a working prototype with an AI coding agent. Product direction came from me; implementation and technical documentation were AI-assisted. This submission text was also prepared with AI at my request. I am not claiming independent coding proficiency or completed physical-iPhone testing. Camera, location and voice behaviour still need real-device validation.
Next steps
Test real packaging on an iPhone, expand documented menu coverage and improve accessibility. Affiliate links and open-ended voice chat are future possibilities, not implemented claims.
Run and inspect
The public repository includes a complete source ZIP, root MIT licence, planning documents and local run instructions. Extract the source and run python -m http.server 4173 --directory dist. An English testing guide is included at dist/guide-en.html.
Log in or sign up for Devpost to join the conversation.