ShiftAI - Dynamic Staffing and Scheduling Agent

We built ShiftIQ because retail managers waste 35 hours weekly on manual scheduling, costing $50K/year in bad shifts. The real damage comes from overstaffing, understaffing, and putting the wrong person on the wrong task. The system works in three stages. Upload employee profiles and historical tasks - we embed them into Qdrant using sentence transformers (skills, certs, performance become semantic vectors). Create new tasks - Gemini 2.0 Flash analyzes complexity and estimates duration with confidence scores. Get matched employees - Qdrant's semantic search ranks candidates by skills, availability, and past performance. Opus workflows orchestrate the entire pipeline from intake to delivery. Tech stack: FastAPI backend with SQLAlchemy for persistence, Qdrant Cloud for vector search, Gemini 2.0 Flash for real-time analysis, Next.js 16 frontend with shadcn/ui, and Opus for workflow orchestration. Deploys to Railway (backend) and Vercel (frontend) with graceful degradation if services are down. What's different: the AI understands context. Semantic search catches patterns like "great at customer service but burns out on consecutive phone shifts." Gemini flags missing certifications or complexity mismatches before they're problems. Opus ensures reliable execution without manual babysitting. Current state: fully functional MVP with file uploads, task creation, AI matching, and schedule generation. Next steps: calendar export, mobile notifications, and multi-location sync for franchises.