top of page
Co-Organisers:
I-RICE'26
4th International Research, Innovation, Creativity & Engineering Project Competition 2026
.png){kind=small}
Project ID:
ITCS04306
VibeFlow: AI-Powered Unified Social Media and Cross-Platform Management Platform
Project Title:
Category:
Information Technology/ Computer Science/ AI
Inventors:
TAN BING YAN, Nur Shamilla Binti Selamat, Noor Fatihah Bt Mazlam, Khairunnisha Binti Ismail
Institution/Company:
Southern University College
Invention Description/ Abstract:
Social media has become the primary space in which Malaysian adolescents and young adults socialise, yet the same platforms expose them to violent, sexual and fraudulent material, and each platform moderates content behind an opaque process offering little visibility and no meaningful right of reply. Two gaps follow. Young users must manage several disconnected applications to maintain one social presence, and they cannot tell why a post was removed, nor whether what they are reading was produced by a machine. This study designs, implements and evaluates VibeFlow, a unified mobile social application that aggregates in-app, Facebook and X content into a single feed and screens every post through two parallel and independently governed artificial intelligence pipelines: one that judges whether content is harmful, and one that judges whether content is synthetically generated. Each pipeline exposes its verdict to the author as a visible label carrying a bounded right of appeal. The system was built over an Agile cycle as a three-tier architecture, specifically a Flutter client, a moderation and community server, and a third-party integration server. Harmful-content screening combines an explicit-image classifier, a speech transcription model and a multimodal policy classifier. Synthetic-content labelling combines a frame-level deepfake detector with a large language model applied to captions, averaging frame probabilities before fusing them with the textual verdict. The system was then evaluated through user acceptance testing with 60 active social media users, of whom 81.7 per cent were aged 34 or below and 95.0 per cent used social media at least daily. Fifteen statements covering the six feature components were rated on a five-point scale. Every component scored above 4.6 on a five-point scale, ranked wallet 4.97, chat 4.89, live streaming 4.83, gifting 4.79, unified feed 4.77 and content moderation 4.64. The weakest item in the entire study concerned safe posts being wrongly blocked, indicating that over-blocking rather than under-detection is the principal barrier to user trust in automated moderation. Visible labelling and a user-facing appeal route are therefore necessary companions to automated screening, not optional additions to it.
Invention Technical Description
System Architecture. VibeFlow is a Flutter-based mobile platform that combines in-app posts with Facebook and X content into a unified timeline. It uses a three-tier architecture: a Flutter client for profiles, chat, live video and payments; a community server for business logic and policy enforcement; and an integration server for retrieving external social-network data. This enables consistent moderation across different media sources.
Harmful-Content Screening. Every post passes through a harmful-content pipeline. Images are analysed using classifiers to detect nudity, gore and violence. Audio and video are processed through speech-to-text, followed by text-policy classification for hate, harassment, self-harm and other harmful speech. A multimodal model can combine visual and textual signals to identify violations. Flagged content is annotated before delivery.
Synthetic-Content Detection. In parallel, VibeFlow detects potentially AI-generated media. Video frames are analysed using a deepfake-detection CNN, with frame-level probabilities aggregated into a video score. Captions are also analysed for machine-generated patterns. These visual and textual signals are combined into a synthetic-content score, with high-confidence cases labelled “AI-generated.”
Transparency & Appeals. Moderation results are presented through visible labels and brief explanations, such as “Removed: Adult Content” or “Flagged: AI-generated.” Users can appeal flagged decisions and provide additional context. Altered or AI-generated content is also clearly identified.
User Evaluation. VibeFlow was developed using an agile approach and evaluated by 60 active social-media users, with 82% under age 35. Participants tested the platform and rated its features on a 5-point scale. All components scored at least 4.6: wallet (4.97), chat (4.89), live streaming (4.83), gifting (4.79), unified feed (4.77), and moderation (4.64). The lower moderation score was mainly associated with false positives, where safe content was incorrectly flagged.
Conclusion. VibeFlow integrates cross-platform feeds with dual AI moderation for harmful and synthetic content. By combining image, audio, video and language analysis with transparent labels and appeals, it improves safety while maintaining user awareness and control. The evaluation shows strong acceptance of its social features, while false positives demonstrate the importance of transparent and appealable moderation.
Demostration/ Presentation Video
Poster/ Broucher/ Invention Photo
Additional Documents
bottom of page