I built an autonomous agent that generates startup ideas.
10 steps. State-driven architecture. ~$1/day in inference costs.
Stack: Python + @pydantic AI/Logfire + Google Gemini + Postgres.
Deep dive on the architecture here: https://t.co/5ir6ykPL1N
#BuildInPublic#LLM
@sudoingX Probably my new monitor (LG 45 inch 5K2K OLED, curved). Great image quality both for work and gaming / media, and so much more ergonomic than my previous dual monitor setup. I spend a lot of my day in front of this thing, and it really makes a huge difference.
@eliana_jordan I'll have to give that a shot at some point. Just wonder how much control I'll have to give up over my edits. I'm using DaVinci Resolve right now. Most of my videos have a lot of screen sharing (different apps that need different crops) plus talking head.
Figured I'd share this here as well. Windows Sandbox was super helpful in testing my installer yesterday.
Now I just need to figure out the best way to do this on macOS. I don't think there's anything equivalent. Might have to use a family member's Mac instead.
Can't believe I never heard of Windows Sandbox before. Super convenient for testing my app installer and catching errors like this one, which I pretty much expected (missing MS Visual C++ Redistributable DLLs).
#buildinpublic
@asaio87 I think that's good advice regardless of whether you're building with AI or by hand. That said, it wouldn't hurt to put some thoughts into how to scale later. Don't build it yet, but know how you'd go about it - even if that means hiring someone else at that point.
@AlexFinn@cremieuxrecueil Crazy! I'm guessing they're mostly targeting elderly people? Feels like otherwise few people watch live TV these days. We cut the cord almost 20 years ago, our kids never really experienced live TV.
@ivanfioravanti Nice, great to see you're already getting good use out of your StarLink! How are LLMs as TTRPG GMs these days? Haven't tried that in a while. Good choice of game. Ran a few sessions of BitD when it first came out.
I'm still rediscovering X/Twitter after well over a decade of absence. Lots of great features, such as lists. But is there really no way to deduplicate across lists?
E.g. I follow several #buildinpublic or #indiehackers lists with significant overlap, but just enough unique content to warrant adding them all. Would love to be able to browse a joint feed, so I don't always see the same posts when switching to another list.
@vivoplt My first language was German, but I've always coded in English, since the mid-80s (starting with BASIC on my C64). I had been learning English in school for several years by then. Everything around computers and tech was in English anyway, which really motivated me to learn it.
@siya_twt_ Yes, I think the fundamentals are still important (data structures, algorithms, design patterns), and proficiency in at least one language will be helpful. But the higher level concerns are even more important to learn - such as system architecture.
Can't believe I never heard of Windows Sandbox before. Super convenient for testing my app installer and catching errors like this one, which I pretty much expected (missing MS Visual C++ Redistributable DLLs).
#buildinpublic
@noahkagan One of many reasons I decided on my (perhaps contrarian) strategy to build single-purchase desktop apps (with BYOK for AI).
I still pay for many SaaS apps, though (e.g. Canva). Like you say - it just has to be worth it. Also happily pay for my Obsidian sync subscription.
@trentvondouglas@ivanfioravanti Really? I've never noticed that. Was running multiple parallel Claude Sessions just fine earlier this week when I was camping off the grid in the mountains.
If you're building desktop apps: Do you use Github Releases (for hosting installers) and Github Actions (for performing Windows + macOS builds)? Has it been working well for you? Any other platforms you'd recommend?
I've gotten pretty good results with Gemma 4, both on my PC (Nvidia 4090) and MacBook M4 Pro (48 GB unified memory). There's a suitable Gemma 4 variant for most hardware, even phones. For coding, Qwen3-Coder seems decent. I hear great things about Kimi and GLM but haven't tried those yet. I'm slightly tempted to buy a Mac Mini M5 whenever it comes out.