Projects

Designed and led a large-scale, standardized migration platform that consolidated production data across Amazon's FinTech organization — moving 9 relational database clusters and 2 data warehouse clusters across 11 AWS accounts and multiple regions into a governed target environment. Built on AWS Database Migration Service (DMS) with full-load plus change-data-capture (CDC) replication for continuous, low-downtime cutover, backed by automated row-level data validation to guarantee parity between source and target.

Highlights

• Migrated a large-scale production estate — 9 relational database clusters and 2 data warehouse clusters across 11 AWS accounts and multiple regions.
• Replaced ad-hoc, manual migrations with a repeatable, auditable framework for cross-account, cross-region data movement.
• Used AWS DMS full-load + CDC to replicate continuously and keep cutover downtime minimal.
• Automated row-level data validation to verify source-to-target parity and catch drift before cutover.
• Added CloudWatch alarms to monitor replication lag and task health for production reliability.

Diagnosed and tuned large-scale Apache Spark jobs running on Amazon EMR that were slow and expensive at scale. By profiling the workloads and fixing the real bottlenecks — data skew, excessive shuffles, suboptimal partitioning, and executor/memory configuration — and right-sizing the EMR clusters, I dramatically cut job runtime and compute cost, saving millions of dollars in annual savings while reducing software delivery latency.

Highlights

• Profiled large-scale Spark jobs to pinpoint bottlenecks — data skew, excessive shuffles, suboptimal partitioning, and memory pressure.
• Tuned partitioning, caching/persistence, join strategies, and executor/memory configuration for major throughput gains.
• Right-sized Amazon EMR clusters (instance types, autoscaling, spot) to cut compute cost.
• Reduced job runtime and cluster cost — delivering millions of dollars in annual savings.
• Improved pipeline reliability and reduced software delivery latency.

Built an MLOps platform that gave the data science team a self-serve lab to train models and run experiments at scale. AWS Batch orchestrates parallel model training and experiment execution; AWS Lambda validates results; S3 + AWS Glue Data Catalog store and catalog historical experiment results so they stay queryable for analysis; and CloudWatch alarms keep the platform reliable in production. The lab supports 10,000+ parallel experiments and model executions.

Highlights

• Delivered a self-serve experimentation lab enabling the data science team to run 10,000+ parallel experiments and model executions.
• Used AWS Batch to fan out model training and experiment execution across many compute jobs in parallel.
• Validated experiment outputs automatically with AWS Lambda before results were persisted.
• Stored and cataloged historical experiment results in S3 + AWS Glue Data Catalog, keeping them queryable for analysis.
• Monitored platform health with CloudWatch alarms for production reliability.

A multi-agent harness, powered by Claude Code, where a specialized crew of eight AI agents builds software collaboratively end to end — a manager that assigns and organizes the work, plus a system designer, database developer, software engineer, test engineer, deployment engineer, code reviewer, and an AI-optimizer that tunes the crew after every iteration. Each agent is scoped to least-privilege permissions and a curated set of MCP tools. The crew runs roughly 99% of day-to-day development, design, testing, and deployment autonomously, while I stay the human-in-the-loop — monitoring the harness and reviewing code and design at critical checkpoints.

The Crew

• Manager — assigns tasks and organizes the workflow across the crew.
• System Designer — designs new projects and features for existing ones.
• Database Developer — data modeling and SQL development.
• Software Engineer — API development and software product build-out.
• Test Engineer — writes unit and integration tests for every engineer's work.
• Deployment Engineer — AWS cloud deployment and AWS CDK infrastructure.
• Code Reviewer — reviews every developer's code and raises CRs / PRs.
• AI-Optimizer — after each iteration, tunes the agents and builds new MCP tools.

Highlights

• Orchestrated a crew of 8 specialized Claude Code agents spanning management, design, database, software, test, deployment, code review, and AI optimization.
• A manager agent assigns and organizes work; engineers hand off across design → build → test → deploy.
• Built-in code review on every change (raising CRs / PRs) before it ships.
• An AI-optimizer tunes the agents and creates new MCP tools after each iteration, so the crew improves over time.
• Scoped each agent to least-privilege permissions and a curated set of MCP tools.
• I stay the human-in-the-loop — monitoring the harness and approving code and design at critical checkpoints.