Principal AI Architect — Agentic AI, AWS Bedrock & Broad-Spectrum AI (Remote)

Role Overview

A hands-on AI Architect to design and deliver enterprise-grade AI systems spanning Agentic AI, AWS Bedrock, and the full AI/ML spectrum.

Key Responsibilities

• Agent Core Design: Architect the foundational agent loop — perception, planning, tool selection, execution, self-reflection, and memory — using frameworks such as LangGraph, AutoGen, CrewAI, or custom Bedrock-native implementations. Design for goal decomposition, error recovery, and dynamic replanning in long-horizon autonomous tasks.
• Agentic Architecture on Bedrock: Build production multi-agent systems using Bedrock Agents — defining supervisor/sub-agent hierarchies, action groups, session memory, and inline agents for complex enterprise workflows.
• RAG & Knowledge Layer: Design Knowledge Base pipelines with vector stores (OpenSearch Serverless, Aurora pgvector, Pinecone) — covering chunking strategies, embedding models, hybrid search, and retrieval optimization.
• Tool & API Integration: Engineer Lambda-backed action groups (Python/Node.js), API Gateway, and Step Functions to enable agents to invoke enterprise APIs, databases, and third-party systems.
• Broad-Spectrum AI/ML: Architect and oversee the full AI lifecycle — classical ML, NLP pipelines, computer vision, fine-tuning, and custom model training on SageMaker.
• Model & FM Strategy: Drive selection across Anthropic Claude, Amazon Titan, Meta Llama, Mistral, and open-source LLMs — evaluating latency, cost, context fit, and reasoning fidelity. Enforce Bedrock Guardrails and responsible AI controls at scale.
• MLOps & Observability: Define CI/CD pipelines (CDK/CloudFormation) for agent and model deployment; monitor via CloudWatch, Bedrock invocation logs, SageMaker Model Monitor, and X-Ray for cost, drift, and hallucination tracking.
• Hybrid & Multi-Platform AI: Integrate AWS-native AI with Azure OpenAI, Vertex AI, or Hugging Face; bridge Bedrock agents with on-premise RPA platforms (UiPath, Automation Anywhere) via event-driven patterns.
• Executive Advisory: Lead C-suite workshops to define AI strategy, shape build-vs-buy decisions, and quantify ROI across the AI portfolio.

Technical Requirements

• Agent Core & LLM Patterns: Deep hands-on expertise in agentic reasoning loops (ReAct, Reflexion, Plan-and-Execute), multi-agent orchestration frameworks (LangGraph, AutoGen, CrewAI), prompt engineering, tool-use design, and context window management for long-running autonomous workflows.
• AWS Bedrock & Agentic Stack: Proficient with Bedrock Agents, Knowledge Bases, Guardrails, Model Evaluation, and inline agent patterns; experienced in configuring supervisor/sub-agent architectures at enterprise scale.
• Broad AI/ML Stack: Hands-on with SageMaker (training, inference, pipelines) and core ML/NLP/CV frameworks — scikit-learn, HuggingFace Transformers, PyTorch/TensorFlow.
• Data & Integration Layer: Strong command of vector DBs (OpenSearch Serverless, Aurora pgvector, Pinecone), RAG pipeline design, AWS Lambda/API Gateway/Step Functions, and event-driven architectures.
• MLOps & IaC: AWS CDK/CloudFormation for repeatable deployments; MLflow or SageMaker Experiments for model tracking; CI/CD tooling for agent versioning and rollback.

Preferred

• Multi-cloud AI exposure (Azure OpenAI, Vertex AI)
• RPA platforms (UiPath, Automation Anywhere)
• AWS Certified Solutions Architect – Professional or ML Specialty