Principal AI Architect

Job Title: Principal AI Architect

Location: Atlanta GA

Work Mode: 5 days onsite (Let me know if anybody asks for 3-4 days)

AI Architect — Conversational & Agentic AI

Job Description

About the role

You will work closely with a client team in Atlanta to design and build a conversational AI product on AWS in a fast-paced environment. You will lead requirement gathering with business and technical stakeholders, own the end-to-end architecture, make fast technical calls, and stay hands-on so the team moves from concept to a production-ready release on an accelerated timeline. You are equally comfortable at a whiteboard with client leadership and in a code review with engineers.

What you will do

•    Lead requirement gathering: run discovery workshops with business and technical stakeholders, map user journeys and conversation flows, capture functional and non-functional requirements, and turn them into a prioritised backlog with clear acceptance criteria.

•    Define the target architecture for the product early and decisively — conversational AI, RAG and agents on AWS Bedrock/Bedrock AgentCore.

•    Design chatbots and assistants that can serve thousands of concurrent users with predictable latency, cost and uptime.

•    Architect agent ecosystems using MCP for tool and data access and A2A for agent-to-agent collaboration.

•    Drive rapid, iterative delivery: get to a working MVP quickly, then harden, load test and take the product to production readiness.

•    Build hands-on alongside the team: agent code, prompts, retrieval pipelines, integrations with data platforms and enterprise systems, and infrastructure as code.

•    Set standards for LLMOps: evaluation, prompt and model versioning, observability, guardrails and cost governance.

•    Own non-functional design: security, identity and access, PII handling, compliance, resilience and disaster recovery.

•    Work as part of the client team: shape scope and trade-offs with stakeholders, run weekly demos and present architecture and progress to client leadership.

•    Run design reviews, mentor engineers and hand over a documented, operable platform (runbooks, architecture decisions, cost model) at the end of the engagement.

Must have

•    Enterprise delivery: architected and delivered multiple production-grade GenAI or conversational AI products end to end, including at least one taken from concept to production on a tight timeline.

•    Embedded, fast-paced delivery: worked as part of consulting or client teams; comfortable with ambiguous scope, weekly demos and senior stakeholder exposure.

•    Requirements and discovery: led discovery and requirement workshops for AI products; able to translate business goals into use cases, conversation flows, user stories and measurable success criteria.

•    Hands-on builder: writes production Python and infrastructure as code (CDK, Terraform or CloudFormation) — not a diagram-only architect.

•    Conversational AI: production chatbots or virtual assistants with multi-turn context, intent handling, session memory, human handoff and multichannel delivery (web, mobile, messaging, contact centre).

•    Chatbot scale: designed systems running at high concurrency in production. Candidates should state peak concurrent users, p95 latency and daily conversation volume they handled.

•    Scale engineering: response streaming, provisioned throughput and quota planning, semantic and response caching, load testing, autoscaling and graceful degradation under model rate limits.

•    RAG: retrieval pipeline design — chunking, embeddings, hybrid search, reranking, metadata filtering, vector stores (OpenSearch, pgvector, Aurora, Pinecone or similar) and groundedness evaluation.

•    AWS Bedrock (essential): foundation model selection, Knowledge Bases, Guardrails, Agents, model evaluation and cost optimisation.

•    Bedrock AgentCore (essential): Runtime, Memory, Gateway, Identity and Observability for deploying and operating agents securely at scale.

•    MCP: designed MCP servers and clients that expose enterprise APIs and data as governed tools, including authentication and authorisation.

•    A2A and multi-agent: orchestration patterns (supervisor, hierarchical, peer-to-peer) using A2A and frameworks such as Strands Agents, LangGraph or CrewAI.

•    Foundations: strong AWS architecture (serverless, containers, networking, IAM, security), distributed systems and API design; Python hands-on.

•    Responsible AI: guardrails, hallucination control, prompt-injection defence, auditability and data privacy in regulated environments.

•    LLM observability and evaluation: Langfuse, Ragas, Bedrock evaluations or similar tools to trace, monitor and evaluate LLM applications.

•    Location: based in or able to relocate to Atlanta; US work authorisation required.

Nice to have

•    Voice AI with Amazon Connect, Lex or speech models.

•    Equivalent platforms on Azure OpenAI, Vertex AI or Databricks Mosaic AI.

•    Fine-tuning, distillation or small-model deployment for cost and latency.

•    Product mindset: conversational UX design, user feedback loops and A/B testing of prompts or flows.

•    AWS Certification on AI/GenAI

•    Domain experience in Finance