This initiative develops an AI-powered inspection tool that helps SFDA inspectors verify whether food, pharmaceutical, cosmetic, and medical device products comply with SFDA regulations. Inspectors can photograph or scan a product, and the system automatically extracts key details and checks them against official registration databases and safety standards. The tool was developed to reduce manual inspection time, minimise human error, and strengthen public health protection.
Initiative overview
The food and drug regulatory landscape involves thousands of products entering the market across four major categories: food, pharmaceuticals, cosmetics, and medical devices. Inspectors are responsible for verifying that each product is properly registered, contains approved ingredients, carries valid import permissions, and complies with applicable safety standards such as GSO standards approved by the Saudi Food and Drug Authority (SFDA). Traditionally, this verification process has been manual and time-consuming, requiring inspectors to cross-reference product labels against multiple government databases and regulatory documents. This creates bottlenecks during inspections, increases the risk of human error, and limits the number of products that can be verified in a given period.
This initiative addresses these challenges by building an intelligent inspection assistant powered by artificial intelligence. The tool allows inspectors to use a mobile application to capture product information through images or text input. The system then uses AI vision models to automatically extract product details such as the product name, ingredient list, barcode numbers, and manufacturer name. Once extracted, this data is routed to specialised verification services based on the product type. For cosmetics, the system checks ingredient safety and product listing status. For pharmaceuticals, it verifies drug registration and import permissions validity. For food products, a retrieval-augmented generation system compares product information against SFDA standards to assess compliance. For medical devices, the system confirms registration status.
The objectives of this initiative are threefold.
- First, to significantly reduce the time inspectors spend on each product verification, enabling them to process more inspections per day.
- Second, to improve accuracy by eliminating manual data entry and cross-referencing errors.
- Third, to create a centralised audit trail of all verification activities, giving administrators and decision-makers a clear view of inspection patterns, compliance rates, and potential risks across product categories.
Other relevant details
Results, outcomes and impacts: The initiative is expected to deliver measurable improvements across several dimensions of the inspection process. On efficiency, the AI-powered extraction and automated verification pipeline is projected to reduce the average product verification time from approximately 15 to 20 minutes of manual checking down to under 30-60 seconds. This represents an estimated 85 to 90 percent reduction in per-product inspection time, allowing each inspector to verify significantly more products per shift.
On accuracy, the system eliminates manual transcription of product details from labels, which is a common source of errors in traditional inspections. The AI extraction service targets above 95 percent accuracy in extracting product names, ingredients, and other details.
Challenges and lessons learned: A primary challenge was ensuring the AI development team fully understood the product verification mechanisms used by inspectors, which required extensive knowledge transfer sessions with inspector experts. Compliance with SFDA specifications and regulations posed another challenge, as the system needed to accurately reflect complex and frequently updated regulatory requirements across food, drug, cosmetic, and medical device categories. Additionally, ensuring the tool can reliably handle over 2,000 specifications and correctly match each product to its relevant standards required building a robust retrieval system and thorough testing against real inspection scenarios. Key lessons learned include the importance of embedding domain experts within the technical team from the start, building the specification database in a structured way that supports updates, and conducting continuous validation with actual inspectors to ensure accuracy.
