This community is large and it moves fast. That's the good part and it's also the problem: valuable posts get buried, the same questions get re-asked instead of found, and spammers show up wherever there's an audience. These guidelines are what keeps the room worth showing up to. Read them once. You won't need them again, because most of this is what you'd do anyway. New here? Start with Jake's welcome post and the Foundation course. This post is about how we behave, not where to begin. 1. Build in public. Post the thing while it's half working. A half-finished build is more useful to everyone else than the polished writeup you'll never get around to, and you'll get corrected before you've spent a week going the wrong way. 2. Teach what you learn. The day you figure something out is the day you're best at explaining it, because you still remember exactly what confused you. A month later you've forgotten the hard part and your explanation gets worse. If you cracked something this week, that's a post. 3. Ask good questions. Specific beats polite. "How should I structure this?" gets three vague answers. "I have a 40 file client folder, the model keeps loading the wrong context file, here's my CLAUDE.md" gets a real one. Say what you tried, what happened, and what you expected instead. A more in depth guide: https://dontasktoask.com 4. Give credit. If you built on someone's skill, template, folder structure or comment, tag them. It costs you nothing and it's the reason people keep publishing their work here instead of keeping it. A lot of the best material in this community started as somebody's reply on somebody else's post. 5. No spam, no pitching. Sharing a tool you actually use and explaining why you use it is fine. Dropping a link with no context, cold DMing members, or treating the feed as a lead list is not. Networking is genuinely welcome (there's a Connection Hub for exactly that: https://www.skool.com/cliefnotes/new-the-connection-hub-is-live) but networking and pitching are not the same thing, and everyone can tell the difference immediately.

1. Go check out 📚Navigating The Course to see how to get around and what's here. 2. Start with The Foundation. Concepts, folder architecture, prompting framework. Everything else builds on this. 3. Check in at the bottom of each lesson. Polls, discussion posts, other members working through the same stuff. Use them. 4. When you're ready to build real things join in on our Biweekly competitions and win some real cash. ⭐ Competitions Mega Thread 5. If you are wanting to dive into the masterminds, grab all the past templates, artifacts and resources. Upgrade and head into the The Vault for Premium and The Drawing Room (VIP) for VIP 6. Post your work. Ask questions. Help others when you can. What are you here to build?

Okay quick context before the numbers, because they only mean something if you know the setup. One used RTX 3090, 24GB, sitting in my main PC. That's the whole rig for this. House rule I stuck to the entire time, one model resident at a time, nothing else fighting it for VRAM. Everything below is from an actual run with files on disk, no vibes, no "trust me." So the first job was 309 real meeting transcripts that needed structured extraction (knowledge, preferences, patterns, the usual). I gave the whole corpus to a local 35B model called ornith. It chewed through all 309 with zero errors in 4.1 GPU hours, roughly 2.4 million tokens in and half a million out, and spat out 8,354 clean structured extracts. Cost me nothing. No API meter running, no per token anxiety, I just let it run overnight. Here's the part that actually surprised me though. I first tried to run this same corpus through cloud. It refused. The transcripts had pasted third party security content in them and the safety filter kept false flagging it, so the cloud model just would not process the batch. Local had zero problem with it. Same content, one machine says no, my 3090 in the corner says sure. That alone paid for the card in my head. Then embeddings. 21,305 wiki chunks embedded by a tiny 137M embedder in about 16 minutes, 0 failures, around 22 chunks a second. That's what powers hybrid search across 2,769 markdown pages now. Again, $0, again, local. But raw "can it run" isn't the interesting question. The interesting one is what job goes to which model, and here I benched it properly. Two local models (ornith 35B and qwen 3.6 27B) across three harnesses (raw ollama, opencode, jcode) across five task types (classify, code, extraction, research, writing). 102 cells total, each one scored against real fixtures, pytest for code, field matching for extraction, not me eyeballing outputs. Biggest surprise of the whole thing, ornith 35B is a genuinely perfect coder. 9 out of 9 code fixtures at a full 1.00, on every single harness. For a 35B model you can run at home, that is not what I expected going in. Extraction turned out to be basically solved too, everything scored above 0.88 so you just pick the fastest one (ornith raw, 10 seconds).

So I've been obsessed with creating interactive art experiences for awhile (though selling them is an entire different story) this one was inspired by Teamlabs Sketch Aquarium and I decided to make my own version. I have also made a dinosaur version (hopefully can sell it to a local dinosaur museum here in town) and a 3d plane version. This works with digital or real life coloring pages you can scan in. My favorite part is it works across the world so you can see other people color their Pokemon and see them swim around! If you want to try it out you can here: https://draw-a-fish-mu.vercel.app/welcome.html Definitely used ICM for this but largely ran into problems of animating it correctly. The process for both bipedal and quadrupeds could be better still.

I hit my token limit for the first time and it did not feel good lol. I am still loading all my projects into VS Code every time as a single large folder that houses them all... I'm wondering, if I stop doing that and JUST load what I'm working on, if that's going to significantly decrease token usage. I am aware that having it point to one project at a time is better, but I'm wondering if it's going to be substantial or just a little thing. Does that make sense? Thanks! Ruby