FinShield-Barclays
Privacy-first multi-modal AI fraud detector
Overview
FinShield is a privacy-first, AI-powered fraud detection platform, originally built for the Barclays Hackathon. Users can submit suspicious emails, links, images, audio, or videos, and the platform runs multi-modal AI analysis to flag likely fraud or phishing attempts — aimed at giving people a fast, accessible way to check something suspicious before they act on it.
Technical Architecture & Specifications
Key Features & Implementation Highlights
- Multi-modal fraud analysis — accepts suspicious emails, links, images, audio, or video for review
- AI-driven detection and explanation of likely fraud/phishing indicators
- Privacy-first handling of submitted content
Challenges & Tradeoffs
The central challenge was handling multi-modal inputs (text, images, audio, video) through a single analysis pipeline while keeping user-submitted data private. FastAPI was chosen over Django for the backend because of its native async support, which is important when each fraud check involves multiple I/O-bound AI inference calls that can run concurrently. The privacy-first design principle meant submitted content could not be persisted or logged beyond the analysis session, which simplifies data retention compliance but makes it impossible to retrain or improve the detection models from production data — a deliberate tradeoff favoring user trust over model improvement.