DerivGuard

DerivGuard is an autonomous oversight nexus for Deriv’s P2P and Affiliate ecosystems. By deploying LangGraph agents to analyze video evidence and audit complex commissions, it eliminates operational bottlenecks and prevents revenue leakage. Long Description (what is your idea?) DerivGuard represents a paradigm shift for Deriv’s financial operations, moving from manual processing to autonomous agentic oversight. Designed to target the "Deep Back Office," this project specifically tackles the twin friction points of P2P dispute resolution and Affiliate commission reconciliation. The system features two specialized sub-graphs orchestrated by LangGraph. The first, the "P2P Sentinel," automates the labor-intensive review of dispute video evidence. Using Vision-Language Models (VLMs), it forensically verifies transaction details, liveness, and timestamps, slashing resolution times from hours to under two minutes. Running in parallel, the "Affiliate Auditor" acts as a "Shadow Comptroller." It independently recalculates complex commission payouts using a decoupled logic engine, instantly flagging revenue leakage, "churning" fraud, or calculation errors that traditional spot-checks often miss. Crucially, DerivGuard prioritizes safety through a "Human-in-the-Loop" (HITL) architecture. While agents handle the heavy cognitive load of analysis and verification, final irreversible financial decisions are gated by human approval. This ensures that Deriv can scale its transaction volume without compromising the trust and compliance standards that define its brand