Alex Khazai

Leading the ML platform and computer-vision capability for national-scale satellite imagery within an Australian Government environment. Re-platformed ad-hoc sandbox training onto governed SageMaker pipelines, reducing training cost and time through reusable containers and spot-aware scheduling. Built a satellite-imagery RAG pipeline on Amazon Bedrock with chunking/indexing for geospatial metadata and retrieval tuned to support grounded responses. Governed model and infrastructure practices aligned to ISM/PSPF requirements, including audit-ready lineage and monitoring.

Designed and deployed Retrieval-Augmented Generation (RAG) pipelines on Amazon Bedrock to enable natural language interaction with satellite imagery datasets, generating contextual explanations of visible features (vegetation, water, infrastructure). Built a domain-specific RAG system for GA’s internal knowledge base, leveraging LangChain orchestration with Bedrock-hosted LLMs and embedding models in Amazon OpenSearch.

Leading the design and deployment of advanced MLOps solutions integrated with AWS SageMaker, improving machine learning workflow reliability, governance, and cost-efficiency in support of geoscientific research. Responsibilities include implementing test-driven development, data versioning, secure cloud solutions compliant with Australian government standards, infrastructure as code, MLOps training, and collaboration with Earth Observation teams.