Page 393 - AI for Good Innovate for Impact
P. 393

AI for Good Innovate for Impact



               The AI engine will validate Non-Profit Organization (NPO) names provided in both English
               and Arabic against specific criteria, including duplication, prohibited words, and generating
               alternative suggestions. The validation will leverage advanced Natural Language Processing
               (NLP) models to ensure accurate results [1].                                                         4.4-Productivity

               Environment Requirements and Deployment Process:

               The API is deployed and running via Uvicorn inside a configured Anaconda environment, and
               can be optionally registered as a Windows service (using NSSM).

               Logging is enabled to collect runtime feedback.

               Based on debugging logs and UAT testing, the system successfully:

               •    Connects to the SQL database
               •    Returns similar names
               •    Flags duplicates based on logic
               •    Returns appropriate validation messages
               Key Features:

               1)    Duplication Check:

               •    Verify if the provided NPO name exists in the database in either language (English or
                    Arabic).
               •    Notify the applicant with descriptive validation messages if:
                    –   The English name is already taken.
                    –   The Arabic name is already taken.
                    –   Both names are taken.
               •    Provide a link to the NPO National Registry for public information on existing NPOs.

               2)    Prohibited Words Check:

               •    Validate names against a list of prohibited terms (e.g., names linked to terrorism or
                    culturally sensitive language).
               •    Prevent the user from saving prohibited names and display a descriptive message
                    indicating the issue.

               3)    Alternative Name Suggestions:
               •    Generate alternative name recommendations using an AI model trained on linguistic
                    patterns.
               •    Ensure suggested names comply with business rules.
               •    Enable the user to click on a [Generate Alternative Names] button to view the
                    recommendations.

               The validations are guided by predefined policies which are dynamically updated:
               •    Exact Match Policy: Disallows submission if the name exists exactly (either in English or
                    Arabic).
               •    Similar Name Notification Policy: Flags and informs the user if a semantically or visually
                    similar name exists.
               •    Prohibited Terms Policy: Names associated with terrorist organizations, government
                    bodies, or restricted terms are blocked.



                                                                                                    357
   388   389   390   391   392   393   394   395   396   397   398