Beyond RAG
TAG, CAG, MAG and KAG on a self-hosted GPU stack: what each pattern actually does, what I have built, what I measured, and when to use which.
Blog
Thoughts on Data Engineering, Cloud Architecture, and AI Development.
Deploying ADK Agents
Three ways to deploy a Google ADK agent on GCP with agents-cli: Vertex AI Agent Engine, auto-scaffolded Cloud Run, and a Terraform-managed Cloud Run service.
Athena on AgentCore
Building a read-only Athena assistant with Strands and Amazon Bedrock AgentCore: MCP tools, a Lambda tool, skills, and an external model provider when Bedrock quotas get in the way.
Evaluating ADK Agents
Behavioral evaluation of a BigQuery agent built with Google ADK: criteria, scenarios, judges, and the cost lessons that came with them.
RAG on vllm-gke
Adding retrieval-augmented generation to a self-hosted GPU stack on GKE: Qdrant, CPU embeddings, hybrid search, reranking, and an evaluation harness, all measured before being trusted.
Building a BigQuery Agent with Google ADK
A hands-on guide to provisioning a BigQuery data assistant on Google Cloud using the Agent Development Kit (ADK) SDK, featuring MCP tools and skill integrations.
Multi-Tier LLM Routing on GKE
How to serve multiple models behind one endpoint and route intelligently based on prompt complexity using Envoy AI Gateway and vLLM Semantic Router on GKE.
vLLM Scale-to-Zero on GKE: KEDA Meets Datadog
Monitoring and scaling a vLLM stack on Google Kubernetes Engine with Datadog, using a custom proxy for synchronous scale-to-zero.
From DataZone to SageMaker Unified Studio: A Cataloging Evolution
A high-level overview of the transition from Amazon DataZone to SageMaker Unified Studio, exploring the structural changes and migration impact.
AWS MCP Server as Widgets
Connecting a generative UI host to the real AWS MCP Server: CORS proxying, OAuth 2.1 sign-in, and live AWS data rendered as widgets.
Scaling vLLM to Zero with KEDA on GKE
A deep dive into reducing idle costs by scaling a vLLM GPU node to zero using KEDA and its HTTP add-on in GKE.
SageMaker Unified Studio Cross-Account and Cross-Region Data Sharing
Deep dive into project creation with the Lakehouse Database blueprint across accounts and regions from a single SageMaker Unified Studio domain.
MCP Apps: When the AI Chat Ships Its Own UI
Exploring how AI chats generate dynamic user interfaces using MCP Apps and the BYOK model.
From DataZone to SageMaker Unified Studio: Datalakehouse Blueprint Evolution
Deep dive into the DataLakehouse Blueprint differences between DataZone and SageMaker Unified Studio (SMUS) and how to reproduce the v1 layout.
Deploying an Enterprise vLLM Stack on GKE
A practical guide to deploying and serving open-weight models on Google Kubernetes Engine using vLLM for agentic coding workflows.
SageMaker Unified Studio Projects
A deep dive into the architecture and lifecycle of projects in SageMaker Unified Studio.
WebMCP: The Invisible Layer for AI Agents
A deep dive into the Model Context Protocol (MCP) and building WebMCP tools for AI agents.
From Amazon DataZone to SageMaker Unified Studio: A Strategic Guide to Blueprints
A deep dive into AWS-managed and custom blueprints in SageMaker Unified Studio.
Obsidian Power Setup: Git & AI
How to securely sync your Obsidian vault with Git and integrate AI features via Copilot for a powerful second brain.
Hello World — This Blog is Live
Welcome to my blog about Data Engineering, Cloud Architecture, and AI Development.