Staff AI Platform Engineer: Agent & Retrieval Infrastructure

Job description

Job responsibilities

• Architect Agent Orchestration: Design the Amazon Bedrock integration, including agent and action group configuration, backend APIs, model access, throughput, and cross-environment deployment.
• Manage Retrieval Data Plane: Own the end-to-end retrieval pipeline from ingestion and chunking to embedding and storage in Amazon OpenSearch Serverless. Focus on optimizing for index design, cost, and capacity.
• Extend Data Pipelines: Adapt ingestion pipelines for internal knowledge, ocean data, and customer platforms, addressing challenges specific to geospatial and large-binary datasets.
• Secure AI Infrastructure: Implement robust security including Bedrock Guardrails, VPC and PrivateLink network boundaries, least-privilege IAM, and audit trails to ensure data isolation.
• Define Agent Governance: Build the mechanisms to enforce approval boundaries for autonomous actions, ensuring agents are safe and monitored.
• Establish LLMOps & Observability: Implement comprehensive monitoring for tracing, tool calls, and retrieval performance, using CloudWatch and LLM-specific tools like Langfuse or Phoenix.
• Build Evaluation Frameworks: Create the infrastructure to run automated evaluations, track results, and manage release gates for model accuracy.
• Enable Engineering Productivity: Provide the team with abstraction layers, SDKs, and self-service environments that allow engineers to ship AI features independently.

Job requirements

• 8+ years in software and infrastructure engineering, including deep production backend experience (Python or TypeScript preferred, Go fine) and staff-level ownership of technical direction.
• Hands-on experience standing up Amazon Bedrock in production: agents, knowledge bases, guardrails, model access, and the throughput and quota decisions that come with them.
• Containerized service deployment on ECS, EKS, or Lambda, with CI/CD you have owned rather than inherited. The models are managed, but the backend APIs, tool endpoints, and ingestion jobs still run somewhere real.
• Practical RAG and vector search experience: embeddings, chunking strategies, semantic search quality, and operating a managed vector database (OpenSearch Serverless, Pinecone, pgvector, or similar) at production scale…
• Real data engineering: you have built or substantially extended ingestion pipelines over messy, heterogeneous, unstructured sources, and you think about freshness and correctness as SLAs rather than afterthoughts.
• Strong AWS ecosystem expertise: IAM roles and least privilege for machine identities, VPC networking and PrivateLink, Lambda, S3, KMS, CloudWatch, and provisioning safely through infrastructure as code (Terraform, CDK,…
• Production LLM exposure: you have moved LLM features or autonomous agents past the prototype stage into environments other people depend on.
• A working point of view on securing agentic systems: scoping tool permissions, prompt injection and exfiltration risk, sensitive data handling in retrieval, and where a human belongs in the loop.