--- title: 'POD Combined' description: "Built it three times. Thought the problem was prompting. It wasn't." intro: "I rebuilt this three times before it worked. The fix wasn't better code, it was better constraints." outcome: '305 users in first 2 months. Built solo in 3 weeks.' date: '2025-03-01' url: 'https://www.podcombined.com' coverImage: '/images/pod-thumb.png' ogImage: '/images/pod-og.jpg' --- ## The Product Print-on-demand sellers manually compare profit margins across providers using spreadsheets. No side-by-side comparison tool existed. POD Combined delivers real profit margins across 100+ provider-product-region combinations in under 30 seconds. CSV over database, USD-only storage — prove the calculation engine first, invest in complexity later. ![POD Combined — outcome](/images/pod-img-2.png) ![POD Combined — outcome](/images/pod-img-3.png) ## The Process The product story isn't the interesting part. The process story is. This was my first attempt building end-to-end with AI as co-pilot. I rebuilt it three times before finding the right approach. - **Build 1 failed:** prompting reactively, fixing problems as they appeared. AI over-engineered and missed edge cases. - **Build 2 improved:** realised better prompting wasn't enough. Created a structured Knowledge Base for context and constraints. - **Build 3 worked:** invested entirely in the Knowledge Base before writing code: project overview, product laws, tech stack, development phases. The AI stayed on track because the constraints were clear. > That Knowledge Base approach became my foundation for everything I built after. AI handled backend logic and deployments, but the UI needed to feel human. That's where I stay hands-on. ![POD Combined — outcome detail](/images/pod-img-4.png) ## Outcome 305 users in the first 2 months, with sellers using it daily for margin calculations. A clean, live product that works. > Distribution is the next challenge.