面議(經常性薪資達4萬元或以上) 桃園市龜山區 5年工作經驗 2天前更新
【Job Overview】
We are seeking an expert AI Platform & Agent Architect to lead the design, implementation, and operationalization of GPU-based LLM infrastructure, secure AI Gateway systems, and orchestrated multi-agent workflows. This role entails building AI Gateways to route and govern model/tool access, operating MCP servers, enabling multi-agent collaboration, designing RAG pipelines, and establishing robust LLMOps/MLOps practices.
【Responsibilities】
1. Design and manage GPU-powered LLM inference platforms (e.g. vLLM, Triton) on Kubernetes with CI/CD, monitoring, and resource optimization.
2. Architect and deploy a scalable AI Gateway layer for secure, efficient, fallback-capable model and tool access with observability and rate limiting.
3. Develop and maintain MCP servers, including defining tools/resources, implementing JSON‑RPC session flows and context negotiation per MCP specs.
4. Build multi‑agent orchestration frameworks using MCP, coordinating context exchange, tool usage, session management, and shared memory.
5. Design and implement RAG pipelines using vector databases ( e.g., FAISS, Qdrant, Chroma) to augment agent workflows.
6. Establish LLMOps & MLOps best practices: model versioning, GPU utilization tracking, CI/CD pipelines, and observability (leveraging OpenTelemetry for metrics, traces, and logs).
7. Integrate enterprise-grade authentication (SSO/OAuth) and enforce governance, context-level access control across Gateway and MCP layers.
【Basic Qualifications】
1. Proven experience building AI Gateways: secure routing, token management, failover, rate limit, and observability.
2. Hands-on with MCP server/client architecture: context specification, JSON-RPC session handling, tool negotiation.
3. Strong track record designing multi-agent systems with coordinated tool use and shared context delivery.
4. Expert knowledge in GPU-based LLM inference and orchestration (Kubernetes, CI/CD, inference engines).
5. Experience building Retrieval-Augmented Generation pipelines and integrating vector databases.
6. Solid LLMOps/MLOps background—model lifecycle management, observability, deployment automation.
7. Excellent collaboration and communication skills for cross-functional system architecture work.
【Preferred Qualifications】
1. Familiarity with security and governance for MCP/gateway layers: session scoping, prompt injection mitigation, permission modeling.
2. Proven skills in prompt engineering, model fine-tuning, and inference optimization.
3. Prior experience building internal AI platforms or model registries.
展開 年節獎金年終獎金績效獎金員工團保眷屬團保