Fractional AI Data Architect
Specialized fractional architecture for organizations whose competitive advantage will be how they harvest, structure, and reason over their data. Iceberg, Bedrock, Claude, agentic systems — built deliberately.
The right moment.
Engagements are most useful when the moment is right. Below are the situations where this kind of work has produced the most value — for the client, and for the architectures that have followed.
You're committing to AWS as your data platform and want senior architecture from the start.
You're building agentic AI systems that need to query enterprise data — and the stakes for getting the data layer right are high.
You have an existing data lake that has accumulated decisions and you need someone who can see what's load-bearing and what's not.
You're integrating Bedrock, Claude, or other LLMs into operational workflows and want a designer-in-the-room as the patterns are established.
Tangible deliverables.
Architecture and operational design for AWS-native data lakes — Iceberg, Spark, Glue, Lake Formation, EMR, Redshift, Athena.
Patterns for AI / agentic systems — Bedrock orchestration, retrieval architectures, schema-aware context, multi-agent designs.
Adaptive ingestion design — schema-drift handling, automatic relationalization, source-system change isolation.
Cost-allocation models that connect engineering posture to run-rate dollars.
Working code examples and reference implementations when the conversation needs them.
Engagement structure.
Architecture working session
An initial deep technical conversation to map the current state and the strategic ambition. Half-day, on-site or remote.
Recurring weekly cadence
One to two days per week of focused work — architecture, design reviews, code-level guidance, agent design, vendor selection.
Production transitions
Decisions are documented. Implementations are reviewed. The team retains the architecture once the engagement concludes.
Where this work has lived.
Designed and built a consolidated enterprise data lake on AWS EMR and Apache Iceberg, with a Python/Spark auto-relationalization engine that turns complex on-premises Workday and PeopleSoft data into queryable relational tables, plus a serverless Student Data Warehouse on Step Functions, EventBridge, and Lambda.
Implemented Bedrock AI agents for automated PII classification at 95%+ accuracy across the data lake, plus an LLM-based pipeline-failure triage system that meaningfully reduced time-to-diagnosis for the operations team.
Start with a conversation.
Reach out with the question on your mind. If this engagement is a fit for your moment, we'll talk about scope and structure.