Agent与大模型安全算法工程师-飞书
Core
Build systematic evaluation datasets and benchmarks for LLM security risks (Prompt Injection, data leakage, hallucinations) and design runtime behavior detection algorithms for AI Agents to ensure safety.
Role type
Senior IC LLM Security & Agent Safety Algorithm Engineer
Builds
Security evaluation standards, risk detection algorithms, and Red Team attack vectors for Feishu's large model and Agent systems
Domain
AI Safety, Large Language Models, Agent Systems
Deliverable
production ML models
Required skills
Python/Go, Transformer architecture, SFT/RLHF/DPO alignment techniques, Prompt Injection defense, Agent tool calling chains, RAG architecture, runtime behavior detection, anomaly detection, time-series modeling
Preferred skills
Red teaming experience, adversarial machine learning research, threat modeling, jailbreak defense, data poisoning mitigation
Technologies
LangChain, AutoGPT, Hadoop, Spark, Flink, Reward Models
Responsibilities
Construct security evaluation datasets and benchmarks for LLM risks; Design Agent runtime behavior feature systems and risk detection algorithms; Research model consistency and stability under attack perturbations; Establish and execute LLM Red Team mechanisms to discover structural weaknesses
Seniority
Senior, hands-on IC