Principal AWS AI Architect (Hybrid) Cincinnati or Atlanta (Norcross)

Principal AWS AI Archiect (Hybrid)

Location: Cincinnati or Atlanta (Norcross)

Interview mode: Virtual

Job Description:

Notes : A hands-on architect who combines deep AWS and Amazon Bedrock expertise, advanced Claude application development, and enterprise-grade AI security with the ability to deliver production solutions.

Job Description: Principal AI Architect AWS, Amazon Bedrock, Claude & AI Security

Role Overview

We are seeking a highly experienced, hands-on Principal AI Architect to lead the design, development, and deployment of secure, scalable, enterprise-grade Generative AI solutions. The ideal candidate will have deep expertise in AWS cloud architecture, Amazon Bedrock, Anthropic Claude models, agentic AI, and enterprise AI security.

This role combines strategic architecture leadership with hands-on engineering. The candidate will define enterprise AI architecture standards, establish development best practices, and lead the transition of AI proofs of concept into production-ready applications and autonomous agents.

The architect will work closely with engineering, security, data, infrastructure, and business teams to identify opportunities for AI-driven transformation, integrate AI capabilities into enterprise applications, and deliver measurable business outcomes.

The ideal candidate is a hands-on AI architect who has built production-grade Claude applications and agents, understands AWS deeply, and can establish secure AI engineering practices across development teams.

Key Responsibilities

1. AWS & Amazon Bedrock Architecture

• Design and implement enterprise AI architectures using AWS and Amazon Bedrock.
• Demonstrate strong hands-on experience with Bedrock foundation models, model invocation, Knowledge Bases, Guardrails, Agents, and model evaluation capabilities.
• Architect scalable, resilient, cost-effective AI platforms using AWS Lambda, Step Functions, API Gateway, S3, IAM, CloudWatch, and other relevant AWS services.
• Design secure data ingestion, retrieval, processing, orchestration, and inference pipelines.
• Establish reusable architecture patterns, reference implementations, and technical standards for enterprise AI adoption.
• Evaluate models based on accuracy, latency, throughput, security, context requirements, and operating cost.

2. Anthropic Claude Development & Engineering Best Practices

• Demonstrate deep, hands-on experience developing production applications using Anthropic Claude models through Amazon Bedrock and, where appropriate, the native Anthropic API.
• Build Claude-powered applications using tool calling, structured outputs, prompt engineering, context management, and multi-step agentic workflows.
• Establish reusable Claude development frameworks, prompt templates, coding standards, model selection guidelines, and architecture patterns.
• Apply advanced techniques for context-window management, token optimization, caching where supported, latency reduction, and cost optimization.
• Implement robust application behavior through schema validation, error handling, retries, timeouts, fallback strategies, and deterministic business-rule enforcement.
• Develop automated evaluation and regression-testing frameworks to measure accuracy, hallucinations, instruction following, and consistency across model or prompt changes.
• Establish development best practices that separate model-generated outputs from trusted business logic and authorized execution.
• Keep current with relevant Claude capabilities, model changes, API features, and emerging agent development patterns.

3. AI Agents & Enterprise Automation

• Design and build AI agents capable of executing multi-step business workflows using tools, APIs, enterprise data, and business applications.
• Implement appropriate agent orchestration patterns, including single-agent and multi-agent architectures where justified by business requirements.
• Integrate AI agents with ERP, CRM, databases, document repositories, and enterprise services.
• Design human-in-the-loop approval mechanisms, authorization checks, exception handling, and escalation workflows.
• Implement reliable execution patterns with bounded retries, idempotency, transaction controls, and recovery mechanisms.
• Ensure agents operate within explicitly defined permissions and cannot independently perform unauthorized business actions.
• Identify opportunities to automate complex enterprise processes and deliver measurable improvements in productivity, accuracy, and cycle time.

4. AI Security, Privacy & Responsible AI

• Define and implement security-by-design standards for LLM applications, Claude-powered agents, RAG pipelines, and enterprise AI platforms.
• Conduct threat modeling for AI applications, including prompt injection, indirect prompt injection, jailbreaks, sensitive-data leakage, insecure tool execution, and excessive agent autonomy.
• Implement least-privilege IAM, workload identity, secrets management, encryption, network controls, and appropriate data access boundaries.
• Implement Amazon Bedrock Guardrails and complementary application-level controls for content filtering, sensitive-data protection, and policy enforcement.
• Secure RAG architectures against unauthorized document retrieval, cross-user data exposure, malicious retrieved content, and access-control bypasses.
• Enforce authorization and business policies outside the LLM, validating every privileged tool call and consequential action independently of model-generated instructions.
• Implement human approval controls for high-impact or sensitive operations, including financial transactions and critical enterprise data changes.
• Establish audit logging, security monitoring, incident response, adversarial testing, AI red teaming, and security regression testing.
• Integrate security checks into the software development lifecycle, including code review, dependency scanning, infrastructure-as-code validation, and CI/CD security gates.
• Partner with enterprise security, privacy, compliance, and risk teams to meet applicable organizational and regulatory requirements.

5. RAG, Knowledge Systems & Enterprise Data

• Design retrieval-augmented generation architectures using embeddings, vector search, metadata filtering, reranking where appropriate, and hybrid retrieval.
• Implement enterprise knowledge assistants grounded in authorized documents, structured data, and business systems.
• Evaluate vector storage and retrieval options based on data scale, relevance, security, operational complexity, and cost.
• Establish document ingestion, chunking, indexing, metadata management, freshness, and access-control strategies.
• Implement source attribution, retrieval-quality evaluation, hallucination reduction, and appropriate abstention behavior when evidence is insufficient.
• Optimize retrieval and context assembly to balance answer quality, latency, and token consumption.

6. Production Engineering, DevSecOps & MLOps/LLMOps

• Own the architecture and technical approach for moving AI proofs of concept into secure, production-ready applications.
• Establish CI/CD pipelines, infrastructure as code, automated testing, environment promotion, and controlled deployment practices.
• Implement observability for model invocations, agent traces, tool calls, latency, errors, token consumption, cost, and application-level outcomes.
• Establish evaluation datasets, quality thresholds, regression testing, canary releases, and rollback strategies.
• Design for availability, scalability, resilience, rate limits, quotas, concurrency, and graceful degradation.
• Establish operational runbooks, incident management, model and prompt versioning, and continuous improvement processes.
• Optimize total cost of ownership while maintaining security, reliability, and acceptable performance.