Original Summary

I’ve been working on this thing for about two years now. It started pretty simple. I wanted to see if I could build a system that could take a much deeper look at games than just season averages, recent box scores, or a handful of obvious stats. That slowly turned into a much bigger project. At this point the system is constantly evaluating player and team performance, matchup data, market information, usage, injuries, recent changes and a bunch of other variables I’ve spent a long time testing and adjusting. I’m keeping some of the actual weighting and decision-making process pretty close to the chest, mostly because that part took me forever to build and it’s kind of the guts of the whole thing. I’ve probably rebuilt parts of it more times than I can count. More recently I added another layer that has made the project a lot more interesting to me. I built separate AI analyst agents that can look at the research the system produces and make their own call on it. I didn’t want them to just take a model prediction and write a paragraph about why it must be right. That felt pointless. The analysts can look through the underlying research, matchup context, injuries, usage, pricing and whatever else is relevant, then decide whether they actually agree with what the system is seeing. Sometimes they agree. Sometimes they don’t. That’s probably the part I’ve had the most fun building. Anyway, the NFL side has had a pretty wild start this season. 62-33-1 through 96 publicly tracked NFL predictions. +26.81 units. 27.1% ROI so far. I know 96 picks isn’t some gigantic sample size, and I’m definitely not trying to act like I solved sports betting. There are going to be bad weeks. There will probably be stretches where I’m staring at the thing wondering what the hell happened. But after spending two years building it, breaking it, changing things, rebuilding pages, dealing with data feeds and second guessing the whole idea a hundred times, it’s honestly pretty damn cool seeing it perform like this in public. One thing I care a lot about is keeping the record honest. The picks are published and then they stay there. Wins, losses, all of it. I don’t want this to turn into one of those sites where all you ever see is a screenshot after a good Sunday. Right now everything is free too. There’s no paid picks section or anything like that. I’m much more interested in getting people to use it, follow it and see how it holds up over a much larger sample. The project is called ATSKillers. https://atskillers.com/ NFL archive if anybody wants to go through the actual record: https://atskillers.com/nfl-archive I’m still building a ton of it and I genuinely don’t know exactly what this ends up becoming yet, but this is probably the first time since I started where I’ve felt like, alright, maybe I actually built something pretty cool here. I've had some haters call it vibecoded slop but the software for generating predictions is real and has been worked on tirelessly. &

  • 情报分类：商业与市场研究
  • 分类依据：内容涉及商业、投资或市场动态
  • 信息来源：Reddit · SaaS
  • 发布时间：2026/10/5 23:56:49