Responsibilities

📍Design “Sherlock”-> 📍Build reasoning-capable agents that analyze user behavior, device fingerprints, and trading history to assess whether withdrawals are fraudulent or safe 📍Develop market manipulation detection agents that analyze unstructured social sentiment alongside order book data 📍Identify pump-and-dump schemes and wash trading 📍Build LLM-based triage systems to handle high volumes of risk alerts 📍Use agents to pre-screen alerts by analyzing alert context and historical false positives 📍Automatically decide whether to close alerts or escalate them to human analysts 📍Reduce alert fatigue for Ops teams by filtering noise with high semantic understanding 📍Use LLMs to extract features from messy, unstructured data such as chat logs, support tickets, KYC documents, and chat groups 📍Convert vague risk signals into structured features (e.g., JSON) for downstream risk models or rule engines 📍Build RAG pipelines that allow agents to query repositories of past fraud cases and identify recurring patterns 📍Develop analyst copilot tools to draft investigation reports (SARs) and summarize complex cases 📍Enable human-in-the-loop workflows where agents gather evidence, propose decisions, and humans review and approve outcomes

Requirements

📍LLM application experience: 2+ years building applications with GPT, Claude, or other open-source LLM models 📍Agent frameworks: Deep familiarity with LangGraph, LangChain, or CrewAI 📍RAG for domain knowledge: Experience building retrieval systems that fetch relevant contextual information 📍Evaluation (crucial): Proven ability to design evaluation sets to measure agent performance, such as correctly identifying fraudsters versus regular users 📍Python mastery: Ability to write production-ready, maintainable code (not just scripts)

Looking for your next challenge? The world of crypto offers exciting roles in blockchain development, web3 innovations, and remote opportunities.