Senior AI Architect

Role summary

We’re hiring a AI Architect to design and guide end-to-end AI solutions—from problem framing and model selection (LLMs/classical ML) to environment architecture, security, and cost performance. You’ll collaborate closely with a Solution Architect (overall system design/enterprise fit), while you own the AI architecture: which models, which vector/search technologies, which AWS services, how to integrate with data platforms, how to run safely and cost-effectively at scale.

What you’ll do

• Own AI architecture for initiatives (RAG, agents, predictive models, NLP, vision): define reference architectures, target state, and patterns (batch/real-time, online/offline inference).
• Model selection & evaluation: choose LLMs/foundation models by use case (Bedrock models—Anthropic, Mistral, Meta, Cohere, etc.; SageMaker-hosted custom; open-source). Define evals (quality, latency, safety), guardrails, and fallback strategies.
• AWS solution design (AI/ML stack): map use cases to services such as Amazon Bedrock, SageMaker (Studio/Training/Inference/Experiments/Model Registry), S3/Lake Formation, Kendra (or alternatives), OpenSearch, Lambda/Step Functions, EKS/ECS, API Gateway, CloudWatch/CloudTrail, ECR, Secrets Manager/SSM Parameter Store, KMS, MSK/Kinesis, Glue, Athena, Redshift, PrivateLink/VPC endpoints, WAF.
• Vector & search tech choices: evaluate and standardize options (e.g., Kendra, OpenSearch vector, MongoDB Atlas Vector, pgvector, Pinecone, Weaviate) including ingestion, schema, embeddings, filters, TTL, and ops.
• RAG/Agentic patterns: design retrieval pipelines (chunking, hybrid search, re-ranking), prompt orchestration, tool-use/function-calling, persona/policy layers, caching, and safety filters.
• Environment planning: define dev/test/prod topologies, network isolation, data zones, GPU/accelerator strategy, CI/CD for models (MLOps) and prompts (PromptOps), blue/green or canary rollout for models.
• Cost & performance engineering: produce FinOps projections and guardrails (token/throughput budgeting, autoscaling, spot strategy, quantization, distillation, response caching, batch vs. real-time tradeoffs).
• Security, privacy, and governance: design guardrails for PHI/PII; encryption (at rest/in transit), row-level/column-level controls, key management, data retention; prompt-injection/EOP mitigations; model risk documentation.
• ML platform collaboration: work with data scientists/ML engineers on feature stores, experiment tracking, offline/online parity, A/B tests, and evaluation pipelines.
• Operational readiness: SLOs/SLIs, tracing and telemetry for prompts & models, incident playbooks, reliability and capacity planning.
• Partnering & enablement: co-author solution docs with Solution Architect; create reference implementations, templates, and internal standards; mentor teams.

Required qualifications

• 8+ years in AI/ML and data engineering combined, with 3+ years as an architect for production AI systems at enterprise scale.
• Proven delivery of LLM-powered apps (chatbots/agents/RAG) and classical ML services (forecasting, classification, ranking) in production.
• Deep hands-on AWS experience across Bedrock and/or SageMaker plus core data/compute/networking (S3, Lake Formation, Glue, Redshift/Athena, Lambda/Step Functions, EKS/ECS, VPC, IAM, KMS).
• Strong grasp of vector search and retrieval design, embeddings, re-ranking, and metadata filtering; experience with at least one managed vector/search option (Kendra/OpenSearch/MongoDB Atlas Vector/Pinecone).
• Solid MLOps practices: model registries, CI/CD, automated evaluation, feature stores, model/package versioning, rollbacks, and observability.
• Security & compliance literacy (IAM, tokenization, KMS, network isolation, audit), with experience in regulated data (e.g., PHI/PII).
• Ability to build cost models and optimize end-to-end latency/throughput and spend (GPU sizing, autoscaling, caching, quantization).
• Excellent architecture documentation, stakeholder communication, and leadership with cross-functional teams.