P2p fraud detection and policy enforcement project

Triangulation scams are increasingly becoming inherent in p2p transactions, third party payment frauds and off platform social engineering are putting vendors at risk and businesses at higher reputational risks. Agentic Guard seeks to provide p2p chat security and fraud detection through three distinctive means. Firstly, it examines and analyses chats for inappropriate communication patterns; scans images to detect fraud activities and integrates policy evidence in responses. This is possible through the following pattern: 1. Behavioural Chat Analysis: LLM scans live trades and looks for patterns that reflect fraud activities, it examines keywords such as socia media name use, providing phone numbers in chats, engaging in redirection tactics, among others. it observes this flow and raises alarm to the buyer in the chat while also providing recommendations on best practices 2. Multimodal Evidence Verification: It looks at payment receipts and idnetifies if sender names, amounts match with order details. It also observes fake/original receipts and flags third party payments 3. Policy Implementation: It constantly reminds traders of policies guiding trading conversations by referencing policy standards in warning messages. Tools/Technologies Used Engine: Gemini 1.5 Flash (Multimodal capabilities) Framework: CrewAI & LangChain (Agent orchestration and RAG) Backend: FastAPI (High-performance asynchronous API) Frontend: Streamlit (Real-time security monitoring dashboard)