Writing

—/Writing

Thinking in public.

Notes on what I've learned shipping AI products end-to-end — RAG, prompts, evaluation, and the front-end + accessibility layer that decides whether any of it actually reaches a user.

Forward Deployed9 min read read

Forward-Deployed: Shipping Five AI Systems Zero-to-One

What it actually takes to be the only AI engineer at a brokerage — technical discovery, decomposition, and owning five production systems end-to-end with no handoffs.

Jun 2026

Evaluation8 min read read

An Agent You Can't Measure Is One You Can't Trust

The eval harness — not the model — is what let a production RAG agent survive a GPT-4 to GPT-4o migration and cut factual errors ~60%.

Jun 2026

Agentic Systems7 min read read

When Verification Is the Product

A skiptracing agent the team acts on without re-checking. The contact lookup is easy; the cross-verification that makes it trustworthy is the actual system.

Jun 2026

Agentic Systems8 min read

Underwriting NYC Real Estate in Four Seconds

How I wired NYC's public property registries into a parallel pipeline and a single Claude pass to turn an afternoon of spreadsheet work into a source-cited feasibility report.

Apr 2026

Open Source7 min read

Building an Open-Source MCP Server for AI Coding IDEs

How VIBETTER provides structured, source-grounded intelligence to Claude Code and Cursor using FastMCP 3.0.

Mar 2026

Developer Tools6 min read

Carrying Context Between AI Tools Without Losing the Thread

I kept hitting limits in one AI assistant and starting cold in the next. OmniContext is the local bridge I built so my project state travels with me instead of getting re-explained every time.

Feb 2026

RAG Architecture9 min read

Scaling Enterprise RAG: Lessons from 10,000+ Pages

How systematic evaluation and recursive chunking reduced hallucinations by 60% in a production knowledge retrieval system.

Jan 2026

Agentic Systems8 min read

Multi-Agent Context Orchestration with LangGraph

How Context Refinery turns raw sources into clean, reranked, self-refined LLM context with a four-stage LangGraph pipeline.

Nov 2025

AI Agents6 min read

Building a Multi-Step Reasoning Agent with Tool Use

Architecting an autonomous ReAct agent with GPT-4 function calling, dual memory, and self-correction loops.

Aug 2025