Blog — 25 notes · page 3 of 4
Notes from the build.
Problems I've hit, hurdles I've solved — written down so you don't have to hit them too.
LLM Evaluations and Observability: How to Measure Whether Your AI Application Actually Works
Learn how to evaluate and monitor LLM applications with datasets, deterministic checks, human review, LLM judges, trajectory scoring, tracing, production metrics, and regression testing.
Shopify and eBay Inventory Sync Without Overselling: Build a Source-of-Truth Ledger
A production architecture for synchronizing Shopify and eBay inventory with reservations, idempotent writes, SKU mapping, reconciliation, and safe behavior when two channels sell the last unit at the same time.
How to Build a Production-Ready Voice AI Agent for Customer Support and Appointment Booking
Learn how to design, build, test and deploy a production-ready voice AI agent for customer support and appointment booking, including telephony, realtime audio, tools, security and evaluation.
Text to STEP: why SolidMake couldn't be a GPT wrapper
A language model will happily describe a bracket. It won't hand you geometry a fabricator can cut. Notes on building a text-to-CAD system on DeepCAD and Text2CAD instead of a prompt.
Browser Agents and Computer-Use AI: How to Automate Websites Reliably Without Traditional APIs
Learn how to build secure and reliable browser agents with Playwright, MCP and computer-use models, including architecture, state, authentication, testing, prompt injection defenses and production deployment.
Running AI Locally in 2026: Local LLMs, Quantization, Private RAG and Production Deployment
A practical guide to running large language models locally, covering model selection, hardware, quantization, Ollama, llama.cpp, MLX, private RAG, security, evaluation and hybrid deployment.
List once, sell everywhere — the hard part is the delist
Publishing to five marketplaces is the easy half of multi-channel selling. Everything expensive happens in the seconds after something sells.
n8n, Make, Zapier, GoHighLevel — how I actually choose
Four tools I use every week, and the questions I ask before picking one. Spoiler: the deciding factor is almost never the feature list.