Noon Food competes on catalog quality, but 80%+ of its restaurant partners lack the resources for studio photography. An AI agent orchestrates listing generation and validation — a street stall gets on the shelf on a par with a chain.
Noon Food is the restaurant delivery and q-commerce unit of the Noon ecosystem (UAE / Saudi Arabia / Egypt). It competes with Talabat, Deliveroo, and Careem on lower commission (23% combined vs. approximately 35% at major competitors) and catalog quality.
The partner base is diverse — from chain brands to family shawarma shops and street stalls. The platform has a unified listing guideline (white background, dish-focused composition, readable in a dense feed), but 80%+ of partners lack the resources for photography.
The catalog looked inconsistent: some dishes were on a professional background, others shot from above with a smartphone in poor lighting. The content team manually reviewed every item; some were sent back to the partner for a reshoot, stretching publication time from hours to several days.
The visual codes for Arabic and Middle Eastern cuisine (tableware, garnishes, plating) require their own approach, and off-the-shelf image generation services calibrated for Western food photography did not resonate with the regional audience. Add to that bilingual requirements: every listing must be consistent in both EN and AR simultaneously.
An AI agent orchestrates restaurant listing generation and validation: studio-quality photos from raw shots, bilingual EN/AR validation, and contextual watermarks.
The agent scores each item on 8 parameters (photo-text match, crop, sharpness, lighting, background, EN/AR consistency, foreign objects, mobile readability). The multimodal LLM returns a structured JSON, based on which the item is sent to publication, reshoot, or regeneration.
The agent takes the original photo as a reference for dish composition and texture, then re-photographs it in a studio composition (white background, branded plate, soft shadow, regional plating codes), preserving the actual dish contents. The "regenerate" button works with a single click.
The agent analyzes composition and adaptively applies a watermark (position, opacity matched to frame density), and also detects and removes third-party watermarks from external services before processing.
Input: a list of categories. Output: a set of 200+ covers in a unified seasonal visual style. Previously this task was outsourced to design agencies in cycles of several weeks.
After six months in web interface mode, the agent was integrated into the PIM via API — batch production became a stream: new items are processed without operator involvement, and the validation operator handles only disputed listings.
Content has ceased to be a barrier to joining the platform: a partner without a photography budget gets on the shelf with the content quality of a chain restaurant. This broadened the assortment base in a segment where Talabat and Deliveroo have no equivalent content layer for small players.
The watermark module protected the generated assets — platform content does not leak to third-party apps without a trace, strengthening Noon Food's negotiating position with exclusive partners.
We got a tool that took the most labor-intensive part away from the content team — not the creative part, but the operational one. Previously, onboarding a new restaurant took days and was bottlenecked by the quality of their photography. Now we onboard a partner in one day regardless of whether they have a photographer or not.
The metrics and results presented reflect outcomes of a specific project and depend on its initial conditions. They are provided for informational purposes only, do not constitute a public offer, and do not guarantee similar results in other projects. Supporting materials are available on request.
Generative AI rebuilds SEO content and visuals for product listings: copy, hero images, and rich content tuned to marketplace algorithms — all in a single pipeline.
An AI pipeline assembles lifestyle videos from product photos: decomposition into short artifact-free segments, parallel generation, human in the loop.
A browser agent cross-checks size charts in product listings against official brand websites: auto-search, Vision-based table extraction, and size system normalization.
No cases match the selected filters.
Tell us about your process — we will suggest an audit, pilot, or full deployment.