Role Overview :

We are seeking a highly experienced LLM Systems Architect to design and scale multi-agent enterprise AI systems on GCP. This role focuses on building robust orchestration frameworks, integrating Gemini-based capabilities, and driving production-grade AI architecture across distributed systems.

Key Responsibilities :

- Define and implement multi-agent architecture including orchestrator design, domain agents, request routing, contracts, memory strategies, and tool usage

- Establish scalable system patterns on GKE, including HPA, autoscaling, node pools, and pod disruption budgets

- Design low-latency streaming and messaging architectures (SSE, Pub/Sub, event-driven systems)

- Integrate Gemini (ADK/SDK) with enterprise services and drive migration from legacy REST-based systems where required

- Build robust observability frameworks (tracing, logging, metrics) and evaluation pipelines for LLM systems

- Implement A/B testing, rollout, and rollback strategies for model and agent updates

- Collaborate with security teams to enforce data boundaries, PII protection, prompt-injection safeguards, and access controls

- Create Architecture Decision Records (ADRs), risk assessments, and phased delivery roadmaps aligned with cost, latency, and reliability goals

- Lead architecture reviews and guide engineering teams on best practices for LLM system design

Required Skills & Experience :

- 10+ years of experience in distributed systems and backend architecture

- 2 to 3+ years of hands-on experience building LLM or agentic systems at enterprise scale

- Strong experience with GCP services :

a. GKE, Cloud Run

b. Pub/Sub (event-driven systems)

c. Cloud Logging, Monitoring, Trace

d. Secret Manager

- Hands-on experience with Gemini (ADK/SDK) and LLM integrations

- Expertise in multi-agent orchestration frameworks and system design

- Strong understanding of :

a. Streaming architectures (SSE, real-time pipelines)

b. API design and microservices

c. Proven experience delivering production-grade reference architectures