职位描述

1. 设计车载多 Agent 运行时架构，明确 Agent、Agent Runtime、Model Gateway、Tool Gateway、Safety Gate、SOA Client 和安全监控处理模块的职责边界。
Design the in-vehicle multi-agent runtime architecture and define responsibility boundaries among Agents, Agent Runtime, Model Gateway, Tool Gateway, Safety Gate, SOA Client, and the safety monitoring and handling module.
2. 设计单一共享 LLM、多业务 Agent 共存场景下的任务准入、资源隔离、deadline 保护、推理降级和执行中重评估机制。
Design task admission, resource isolation, deadline protection, inference degradation, and runtime re-evaluation mechanisms for scenarios where multiple agents share a single LLM.
3. 建立 Agent Tool 调用安全框架，包括 Tool Schema、权限控制、Capability Token、Resource Lease、动作仲裁、超时撤权和审计追踪。
Build a safety framework for Agent Tool invocation, including Tool Schema, permission control, Capability Token, Resource Lease, action arbitration, timeout-based revocation, and audit tracing.
4. 定义 Agent Runtime 需要暴露的运行时信号，包括 Agent ID、Task ID、DAG、Tool Call、上下文快照、模型请求、deadline 和协同等待关系。
Define runtime signals exposed by the Agent Runtime, including Agent ID, Task ID, DAG, Tool Call, context snapshot, model request, deadline, and collaboration wait relationships.
5. 支撑车载多 Agent 安全 POC 开发，围绕任务风暴、Agent 死锁、工具链失控、上下文错配、共享模型资源竞争等场景，完成架构设计、机制验证、指标定义和工程闭环。
Support in-vehicle multi-agent safety POC development, covering scenarios such as task storms, agent deadlocks, tool-chain runaway, context mismatch, and shared model resource contention, and deliver architecture design, mechanism validation, metric definition, and engineering closure.

职位要求

1. 本科及以上学历，计算机、软件工程、自动化、电子工程、控制工程或相关专业。
Bachelor’s degree or above in Computer Science, Software Engineering, Automation, Electronic Engineering, Control Engineering or related fields.
2. 具备 LLM 应用或 LLM Infra 实际工程经验，熟悉模型网关、推理服务、Agent Framework、Tool Calling、Function Calling 或 RAG 系统中的至少两类。
Hands-on engineering experience in LLM applications or LLM infrastructure, with familiarity in at least two of the following areas: model gateway, inference service, agent framework, tool calling, function calling, or RAG systems.
3. 熟悉多 Agent 或复杂 Workflow 系统的运行机制，理解任务 DAG、状态机、超时、重试、协同、工具调用和 Trace 追踪。
Familiar with the runtime mechanisms of multi-agent or complex workflow systems, including task DAG, state machine, timeout, retry, collaboration, tool invocation, and trace tracking.
4. 具备较强系统工程能力，能够设计跨进程、跨服务的 Runtime 架构，理解资源隔离、权限控制、服务治理、限流、熔断和降级。
Strong system engineering capability, with the ability to design cross-process and cross-service runtime architectures, and solid understanding of resource isolation, permission control, service governance, rate limiting, circuit breaking, and degradation.
5. 具备扎实的软件开发能力，熟悉 C / C++ / Rust / Python，能够参与 POC 原型设计和关键模块实现。
Strong software development skills, proficient in C / C++ / Rust / Python, and capable of participating in POC prototype design and key module implementation.