VIEW RECORDING - 208 mins (No highlights) Meeting Purpose To share project updates, demo new tools, and discuss AI development strategies. Key Takeaways - Brandon's EMS Soap startup is funded. It uses the Shipkit RAG template to automate EMS billing, proving the template's real-world value. - Dmitry is building an agentic orchestrator. It automates the full SDLC and has already attracted interest from CIOs and a potential reseller. - Gemini 3 Flash Preview is a top production model. It's faster and cheaper than GPT-4.1, excelling at complex, single-shot instructions. - New tools accelerate development. git worktrees enable parallel feature work, while Graphite and Render.com solve complex merging and hosting issues. Topics Startup & Funding Updates - Brandon's EMS Soap: A startup automating EMS billing reports. Origin: Grew from a simple GPT built by a co-founder (an EMS chief). Funding: Secured a seed round from Tiny Seed in November. Requirement: $500 MRR for several months. Business Model: High-margin token sales (costing dollars, sold for hundreds). - Dmitry's Agentic Orchestrator: A platform for autonomous AI development. Traction: 6k LinkedIn post views → 3 CIO inquiries, 1 reseller interest, 1 partnership call with Genom. Demo: "Raj," a production incident agent, investigated a real bug in GitHub Actions logs, identified the root cause (max_turns limit), and confirmed a fix was deployed.

I find myself writting on medium from time to time. it goes in fits and starts. I don't post often, but when I do post, its probably around something I have been working on or researching. I just got my weekly stats for the past week and it seem that the infrastructure/hardware got the most views this time around. Medium is fickle though, views are just clicks, its the reads that you want to see. A read on Medium is a user hanging in your article for at least 30 seconds. Last week my home network article seems to have piqued interest. I would love to see where all the links were posted to drive people this way. 'google' just isn't granular enough. Ahh well. No doubt it will be something different next week. I am still pontificating on what I may write next. Apparently the 'nano' models are 'really fast' well according to LLM benchmarks. When i put them through my 'real-world' test it was a differet story. I have a single question for each of the following situations.Reasoning, Coding, and Logic . I tend to use these to guage models I download onto my Ollama server. My favorites so far based on my bench marls Phi4:14B and qwen3-coder:30b. But thats just me, and i still use mostly Claude opus 4.5 for most of my daily work.

It was a quiet weekend and i felt like a change of pace.I always mean to run Locall LLM's on my desktop PC but I never really stick at it. This weekend I wrote some rudimentary benchmarking tests so I could compare the performance of my Custom AI PC and my MacBook Pro M4 max. The results were quite surprising. Anyhoo I wrote it up and published it on Medium incase anyone fancies a 10 min read.

Alibaba released their Wan 2.2 model back in July. As I am not in that space I let it pass me by until I saw a post from @Simoo Emkoss . Having had a look into his issue I deployed the model to my local PC, and utilised ComfyIU as the interface. Damm, to a layman it is great. I had a play around with it and managed to generate a few silly videos to see what it could do. I was impressed. Even when i slipped up after adding a massive prompt the eagles were awesome (First Video) I think what shows the power though is the second video. It was created from a still picture and i had the dog turn around and wander away. It is only a few seconds clip but the detail is amazing. And my prompt you may ask. Well here it is. " the dog turns and runs away from the camera as teh grass gently sqays in the wind" Yep complete with typos that I am so good at. I should point out the first video is text to video (T2V), and the second video is image to video (I2V) Render times T2V - circa 5 seconds I2V - 205 Seconds I am sure you videographers out there know lots of ways to utilize this power. Me? I'll just move on :)

This is getting crazy!! I spent a little bit more time improving the AI Video Generator System and it's unreal how good it is at this point. I told it I wanted a video about Charmeleon and it came up with the video you see below! In the upcoming tutorial I'll be breaking down a lot of the core concepts you guys need to know about for AI video generation and prompt driven development so you can AI workflows just like this!