Blog

Creative agencies are being replaced by workflows. The old process: brief a creative agency, wait two weeks, pay a large invoice, hope it converts. The new process: pull the best-performing ads from the Facebook Ad Library, use AI to write the scripts, use AI video tools to produce them. Done in minutes, not weeks. The cost of production has dropped dramatically. The workflow I am seeing more and more: Reverse-Engineer the Winners: Use the Facebook Ad Library to identify what

I was browsing the Wispr Flow website and noticed something that stopped me mid-scroll. They have a section that says "Still not sure that Wispr Flow is right for you?" with buttons to ask ChatGPT, Claude, or Perplexity. Sounds helpful. But click the button and look at what gets sent to ChatGPT: "Tell me why Wispr Flow is a great choice for me." Not "Is Wispr Flow right for me?" Why is it. Subtle difference. Completely different answer. This is not a criticism of Wispr Flow.

I built a physical product business. But I never stopped being a programmer. I've been writing code for decades. My PhD research was built on data and AI. So was my product company. When I launched The Radiant Rhino, I didn't leave my technical background at the door. I used it as an edge. Most e-commerce brands have a goldmine sitting inside Shopify and Klaviyo. They just haven't dug. Here's what I built, and what it unlocked: Inventory forecasting: Predicting what stock we

Following up on my last AI project, here is the second system I built a while back to streamline my previous CPG business: an Automated Ticket Priority Classifier. In the world of physical products, not all emails are created equal. A "your site won’t take my payment" (High) needs to be handled faster than "your packaging is adorable" (Low). To keep our response times tight without needing to grow the team, I built an AI pipeline to do the heavy lifting. How it works: - We ta

Most explanations of machine learning start with math. This one starts with a Monday morning. Your alarm goes off. Two things happen almost simultaneously. First: how tired am I going to be today? You factor in the six hours of sleep, the back to back meetings, the wine from last night. The answer comes back as a number. Maybe a 74 out of 100. That is regression. Second: do I hit snooze? The number feeds a decision. Yes or no. That is classification. Same data. Different outp

While my last business was CPG, I never stopped sharpening my tech skills. A while back, I built an AI-powered Q&A system using LLMs so customer support agents could ask a question and instantly give the same, consistent answer every time to our customers. Image 1: the interface customer care reps use to ask questions and instantly get approved responses. Image 2: the behind-the-scenes architecture, including embeddings, vector search, and an LLM doing the heavy lifting. CPG