Writing · Deep dives · status: online

Blog

Long-form technical writeups on AI agents, MCP workflows, and developer platforms.

Make it work. Then keep it working.

Every AI coding demo works. The failures start after: the agent paginates a forty-row table with cursors, sizes a database pool for twenty users at a hundred connections, swallows an error to keep …

Revision: a local-first FSRS-5 desktop app with drag and air gestures

Revision is a local-first desktop app for principal-level interview prep across DSA, system design concepts and use cases, AI concepts and use cases, and behavioral topics, with FSRS-5 scheduling, …

SpendIQ Copilot: Guardrails, Evals, and Honesty for an Agent Over Enterprise Data

Most agent demos are theater: a model, a vector database, and a happy path. I wanted to build the opposite: an agent that touches financial data (vendors, contracts, purchase orders, invoices) and …

We rolled Claude Code out to 20+ enterprise teams in 4 weeks and cut debugging effort 95%

By 2025, every team in our org had someone experimenting with Claude Code. Prompts lived in private dotfiles, MCP servers were duplicated across repos, and nobody measured anything. Adoption stalle…

ServiceNow Table Builder: Overview & Core Capabilities

The ServiceNow Table Builder is an all-in-one, modern low-code developer interface built into the ServiceNow platform. It consolidates multiple scattered data management tasks (creating database sc…

Switchyard: Watch What Happens Inside an Agentic AI + RAG Pipeline

Most RAG tutorials show you the happy path and stop there. But the real learning, the part that makes you useful when a pipeline misbehaves in production, is in what happens around the LLM call: ch…

Putting a Vision-Language Model on a Phone: What 270 Benchmark Runs Taught Me About On-Device AI

Ordo started as a take-home assignment: take a vision-language model (VLM), measure how far quantization can push it, and get it running on a phone. I picked Qwen2-VL-2B-Instruct on llama.cpp and b…

ServiceNow Build Agent Architecture

The ServiceNow Build Agent is an autonomous, AI-driven developer companion integrated natively into the ServiceNow AI Platform. It enables creators and engineers to design, build, test, and deploy …

ServiceNow ALA: Release Lifecycle Documentation AI Agent

The Release Lifecycle Documentation AI Agent is an autonomous governance tool integrated into the ServiceNow platform under the Application Lifecycle Analytics (ALA) framework. Powered by the Now A…

Running Local LLMs on Apple Silicon: Ollama, MLX, and Memory-Efficient Inference

Running serious models locally on a MacBook M1 used to mean one thing: the entire model must fit in RAM. My setup changed when I started combining MLX-optimized models served through Ollama with a …

How We Turned Claude Code Into a Self-Improving AI Engineering Platform for Our Entire Organization

A deep dive into building a shared AI toolkit with 20+ skills, 20+ specialist agents, and safety guardrails - and what happened when we scaled it across 20+ teams in 4 weeks.

ServiceNow Project Portfolio: Technical Writeup

Over my tenure at ServiceNow, I have architected and delivered six major platform capabilities spanning AI-assisted development, low-code/no-code tooling, and internal developer productivity infras…

Technical Specification: Agentic Developer Platform (ADP)

The Agentic Developer Platform (ADP) is a unified, multi-team automation framework designed to bootstrap, govern, and scale AI-assisted engineering workflows across the enterprise. Built and scaled…

ServiceNow Workspace Builder: Overview & Core Capabilities

The ServiceNow Workspace Builder is a dedicated, no-code visualization tool integrated within the ServiceNow platform. It allows citizen developers and business analysts to design, configure, and c…