MLOps Architect - Gen Al

MLOps Platform Architecture

• Design and implement scalable ML and LLM infrastructure on AWS (SageMaker, EKS, S3, IAM, Lambda, Step Functions, CloudWatch).
• Architect end-to-end ML and Generative AI lifecycle workflows:
• Data ingestion & preprocessing
• Feature engineering / embedding generation
• Model training & fine-tuning (traditional ML + foundation models)
• Model evaluation & validation
• Deployment (real-time, batch, streaming)
• Monitoring & retraining

Generative AI & LLM Operationalization

• Architect Retrieval-Augmented Generation (RAG) pipelines including:
• Embedding generation workflows
• Vector database integration
• Document ingestion and chunking strategies
• Retrieval evaluation and monitoring
• Design and deploy LLM-based services using:
• Managed services (e.g., SageMaker endpoints, Bedrock-style APIs)
• Containerized custom inference services
• Establish prompt versioning, evaluation frameworks, and experiment tracking for LLM systems.
• Implement guardrails for hallucination control, safety monitoring, bias detection, and usage logging.
• Define architecture for LLM fine-tuning workflows (including data curation, evaluation, and cost controls).
• Implement scalable orchestration of LLM pipelines using workflow engines and event-driven patterns.

Deployment, Monitoring & Reliability

• Architect scalable inference patterns for:
• Traditional ML models
• LLM APIs
• RAG systems
• Implement model monitoring frameworks for:
• Performance degradation
• Drift detection
• LLM output quality
• Lateness and token usage metrics
• Define SLAs/SLOs for ML and GenAI systems.
• Design safe deployment strategies (blue/green, canary, shadow testing).
• Establish logging, observability, and traceability standards for GenAI systems.

FinOps & Cost Optimization

• Implement cost tracking for:
• Training workloads (GPU utilization)
• Inference endpoints (token consumption)
• Vector database storage
• Optimize LLM workloads for cost-performance tradeoffs (model size, batching, caching strategies).
• Design autoscaling and compute optimization strategies for GPU and CPU-based inference.
• Partner with finance and engineering teams to forecast ML/GenAI infrastructure spend.

Platform Enablement & Standards

• Define enterprise standards for:
• Experiment tracking
• Model registry
• Prompt registry
• Artifact management
• Embedding versioning
• Provide architectural guidance to data science, AI, and engineering teams.
• Evaluate and recommend tooling across the ML/GenAI stack (MLflow, feature stores, vector databases, orchestration tools).
• Drive documentation and reusable patterns for ML and GenAI development.