Portfolio

Featured projects

Shared work. Specific contributions.

Customer-focused diagnostics

CODA - Cosmos Observability and Diagnostic Agent

A local, read-only diagnostic app that turns Cosmos DB for NoSQL configuration, metrics, and optional query logs into recommendations for throughput, partitioning, query efficiency, and cost.

Helps customers understand why their applications are slow or expensive, distinguish capacity shortages from hot partitions and inefficient queries, and evaluate targeted changes instead of simply adding throughput. Estimates savings when usable cost evidence is available.

How it works

A shared Python engine runs deterministic checks, correlates related findings, and preserves the supporting evidence while flagging checks it could not complete. Customers use a local browser UI or connect an AI assistant through MCP; the diagnostic engine itself requires no AI model.

My role

Created CODA because customers often struggle to understand and optimize the performance and cost of applications built on scale-out databases. Its diagnostic capabilities are now shipping to customers in the Cosmos DB VS Code Extension, with broader adoption planned across additional customer experiences.

Preview release. Recommendations depend on available telemetry; incomplete checks are not a clean bill of health. CODA never modifies resources or reads application documents.

CODA browser UI shows a demo Cosmos DB account with diagnostic findings, severity levels, estimated savings, and recommendations for idle throughput and write-heavy throttling. — Example account analysis with demo resource names, recommendations, and workload-dependent savings estimates. Image source

PostgreSQL on Kubernetes

DocumentDB: From Localhost to Multi-Cloud

A hands-on demo of open-source DocumentDB, powered by PostgreSQL on Kubernetes. The DocumentDB operator builds on CloudNativePG to manage primary and replica clusters across Azure Kubernetes Service (AKS) and Amazon Elastic Kubernetes Service (EKS).

Makes database resilience visible through a live travel-booking workload, replication-lag monitoring, and cross-cloud failover. Developers can start locally with Docker and HNSW vector search, then explore how the database behaves across clouds.

In focus: Cross-cloud failover

PostgreSQL write-ahead logs (WAL) stream between clouds over an Istio mesh with mutual TLS. The monitor lets you promote a replica to primary with one click and watch writes move to the new primary. Grafana dashboards expose replication and workload metrics, while replica rebuild controls demonstrate recovery.

My role

Built the demo experience, including the live monitor, booking workload generator, and Grafana observability. Adapted the upstream DocumentDB multi-cloud deployment for an AKS-and-EKS walkthrough.

Built for hands-on learning and live demonstrations; production use requires additional hardening.

DocumentDB Multi-Cloud Monitor shows an Azure primary, Azure and AWS replicas, replication-lag metrics, and controls to promote or rebuild replicas. — Multi-cloud topology with primary and replica health, WAL replication lag, and one-click failover and recovery controls.

Open-source stewardship

Azure Cosmos DB Data Migration Tool

Extensible, cross-platform import and export with configurable sources and sinks.

Gives developers a way to move data between systems and extend the tool for additional formats and sources.

My role

When my team was going to deprecate an engineering owned data migration tool, I worked with a group of community members to create a new one and open source it so others could contribute as well. I launched the open-source tool and am the lead maintainer guiding contributor development.

The command-line core loads source and sink extensions through MEF. Data streams from source to sink as IDataItem records. Extensions implement shared interfaces used by the core. — A modular migration engine with MEF plug-ins, shared contracts, and asynchronous source-to-sink streaming. Image source

Cross-team initiative

Banking Multi-Agent Workshop

Banking scenarios that teach agent collaboration in LangGraph and Microsoft Agent Framework through exercises and reference solutions.

Helps developers work through multi-agent patterns in a concrete domain instead of starting with a blank project.

My role

Initiated the workshop, created all of the deployment infrastructure, heavily contributed to the technical content and provided cross-organizational guidance to continuing contributors.

Banking workshop architecture connects an Angular app on App Service to Agent Framework and LangGraph, Cosmos DB, Azure OpenAI, deployment, and observability services. — Workshop architecture from the LangGraph branch. Image source

Connecting product experiences

Cosmos DB in Fabric Samples

Samples covering CRUD, vector search, Spark, Power BI, reverse ETL, and translytical workflows.

Gives developers concrete ways to explore how operational data, analytics, and AI fit together.

My role

Created the samples gallery, authored most of the samples and guided fellow contributors.

Reverse ETL semantic-search demo shows a natural-language customer query, similar customer result cards, and demonstration metrics. — Reverse ETL semantic-search demo with example customer results. Image source

Extending a shared platform

Travel Multi-Agent Workshop

A build-along travel assistant with persistent agent memory and optional Fabric modules for observability, analytics, optimization, and reverse ETL.

Lets developers explore how agent memory and analytical feedback can inform governed, reversible actions.

My role

Created the workshop/sample experience and all of the infrastructure deployment. Drove the agent analytics and optimization expansion, connecting persistent memory with Fabric insights and governed, reversible actions.

Agent Optimization Dashboard overview displays captured-run metrics and charts comparing models and conversation turns. — Agent analytics and optimization dashboard, showing a demonstration run. Image source

Developer learning

Cosmos DB Design Patterns

Runnable NoSQL data-modeling and architecture examples covering concurrency, partitioning, event-driven designs, and vector search.

Helps developers explore coordination and failure scenarios in working code, making distributed-systems design choices easier to understand.

In focus: Distributed lock

An interactive playground makes lock contention, lease renewal, crash recovery, and fencing-token protection visible through competing workers backed by Cosmos DB.

My role

Initiated this project and was a major contributor, expanding practical examples and improving the developer learning experience.

Distributed-lock playground shows competing workers, lock-holder and renewal controls, protected resources, fencing tokens, and an event timeline. — Distributed-lock playground with competing workers and a live event timeline. Image source

Architecture & deployment

Focused examples that make infrastructure choices and distributed-systems trade-offs easier to explore.

Focused deployment example

Cosmos Web App Network Sample

A public-facing web app connected to Cosmos DB through private networking and managed identity.

Helps developers compare Private Link with service-endpoint and VNet-rule connectivity in a deployable example.

My role

Authored the deployment sample.

Architecture learning · Earlier work

Cosmos Global Distribution Demos

Hands-on demonstrations of consistency, latency, replication, conflict resolution, and failover trade-offs.

Makes distributed-systems behavior easier to reason about through concrete experiments.

My role

Primary contributor to hands-on distributed-systems demonstrations.

Older architecture-learning material targeting earlier .NET versions—not a current production starter.

These projects include work by many people. I've described my own role rather than claiming the whole effort.