PLUTUS AI: An Agentic Mobile Platform for Real-Time Antique & Collectible Valuation

PLUTUS AI is an agentic, AI-powered valuation platform designed to address longstanding inefficiencies in the appraisal of antiques and collectibles. Traditional valuation workflows remain slow, costly, and highly subjective, constrained by fragmented visual data and siloed price histories. This project introduces a mobile-first system that integrates real-time object detection with market-grounded valuation modeling to create a seamless, end-to-end pricing workflow.

The architecture combines a custom-trained YOLOv12 object detection model for rapid item identification with Gemini 2.5 for structured, search-grounded valuation reasoning. Together, these components enable multimodal inference: users scan an item, receive enriched metadata, and obtain a pricing estimate with supporting rationale in under 30 seconds. The system was piloted in collaboration with Rewind Vintage & Consignment to ensure contextual alignment with real-world retail environments. A Supabase-powered backend stores image scans, metadata, and AI-generated valuations, enabling structured outputs and iterative refinement.

Evaluation across catalogued and blind validation items demonstrated strong feasibility, achieving high containment accuracy, clear rationale generation, and low latency response times. While current limitations include dataset scale, visual edge cases (e.g., glare and translucent materials), and reliance on a closed-weight valuation model, the prototype establishes a foundation for scalable, AI-assisted appraisal systems. Future development will focus on dataset expansion, improved visual robustness, open-weight model exploration, and strengthened data governance practices to support ethical and operational scalability.