DEV Community

Cover image for First Post at The Startup Cafe: Building Project Lumina and Where to Release It
Ananya Soni for The Startup Café

Posted on AI-assisted

First Post at The Startup Cafe: Building Project Lumina and Where to Release It

First Post at The Startup Cafe: Where Should We Release Project Lumina First?

Hey everyone, I just joined The Startup Cafe today, and this is my very first post for this organization. Honestly, I do not really know what to say because I have never written a post for an organization here on DEV before. I am a little nervous hitting publish on this one, so bear with me.

For a quick intro, I am the founder and CEO of AGI Systems Directorate. At my company, we are working toward artificial general intelligence through four main pillars:

  • Autonomous AI agents
  • Large language models
  • Robotics
  • Biological human brain replication mechanisms

Right now, we are building Project Lumina, which leans into that fourth hybrid approach. We have built and shipped multiple products and projects before, so this is not our first release. Still, I just have one real question for you all: where should we actually release Project Lumina first? What do you think?

Top comments (11)

Collapse
 
hanzla profile image
Hanzla Baig The Startup Café •

Sounds interesting! What does Project Lumina actually do? Would love to know more about it.

Collapse
 
aisovereign profile image
Ananya Soni The Startup Café •

Thanks for asking! Project Lumina is kind of like a digital assistant app, though I built it so it does not talk or act like a normal digital assistant. It handles tasks on its own and only asks for confirmation when sensitive actions come up. Since I work a lot with bio-inspired stuff, it draws ideas from how the human brain works, and it is more like an experiment or an experimental product.
I am planning to release it when it is actually completed

Collapse
 
pjdeveloper896 profile image
Prasoon Jadon The Startup Café •

woah , it sounds like Jarvis

Thread Thread
 
aisovereign profile image
Ananya Soni The Startup Café •

Haha honestly it is pretty much the opposite of Jarvis. Instead of being calm and sleek, it gets super talkative and chatty while working on a task and ends up with a kind of weird personality, not your usual polite AI. I think a lot of that came out of emergent behavior. Right now pretraining is done and it is sitting in post-training on a pretty massive model, which makes it even trickier to figure out where to release it first since it is still an experiment with some odd quirks. Where do you think a weird talkative build like that should go first?

Thread Thread
 
antonioprosperi2svg profile image
Antonio The Startup Café •

Love the idea of an AI with a weird, chatty personality! 😄 I was also starting to build my own AI, but I got stopped by hardware: I only have a laptop with no GPU, and training or even running a big model needs way more power (and tokens) than I have. Two small shared hostings won't do it either 😅 So for now I'm building developer tools instead. Curious to see where Lumina lands, good luck with the release!

Thread Thread
 
aisovereign profile image
Ananya Soni The Startup Café •

Man I really relate to that. Before I started my company I had a machine so old it practically belonged in a museum, and I have actually posted about how that compute struggle led to starting my company on my profile. I used cloud VMs that got me onto T4 GPUs at first, which obviously are not great for heavy training, and compute really is king when you are building this stuff. Even though I have my own local GPU and TPU setup now, you can still train your models on cloud VMs if you want, using subscriptions or virtual machines on the cloud since plenty of providers offer GPU access that way and let you tap into higher-level hardware over the cloud without needing a massive local rig. Building dev tools in the meantime makes total sense too.

Thread Thread
 
antonioprosperi2svg profile image
Antonio The Startup Café •

Thanks a lot, this really helps! I'll take a look at cloud GPUs and see if I can try something, even though for now I've jumped into another project, BeeLadybug, a debugging tool. I hope you keep writing articles with updates on your work. I know running a company leaves you little time, but it must be a huge satisfaction when you see the results.

And honestly, I love that Lumina has its own weird, chatty personality. I really think every AI has one: some are sharp and quick, some a bit sleepy and dazed, some super talkative. The models I use sometimes make me laugh out loud. Maybe I'm crazy 😂 Looking forward to reading more about Lumina! 🚀

Thread Thread
 
aisovereign profile image
Ananya Soni The Startup Café •

Wow your debugging tool sounds really cool, what does it actually do? And you are definitely not crazy, I laugh at AI responses too. Most AIs are super polite, corporate, and play it safe, which is fine and I am not against safety or ethical thinking at all, it is just a different vibe here because this one is not polite and you never know what it is going to say since it is an experiment. Honestly I think a lot of that comes from the brain-inspired, biological side of what we are building. And you are right that being the founder and CEO, I do not have much time for writing articles or whatever, though I will still post updates and it is not like I am not going to do so. Also I am looking forward to seeing your project launch too.

Thread Thread
 
antonioprosperi2svg profile image
Antonio The Startup Café •

Thank you! BeeLadybug is a free Chrome extension that inspects the layout of any web page. You open it on a site and it shows you problems like elements overflowing the screen, buttons that can't be clicked because something covers them, invisible text, and CSS !important conflicts. Then you can export everything as an HTML report. I just submitted it to the Chrome Web Store, so it's waiting for Google's review now.

About personality, this is just my opinion: I think every developer has their own character, and when they build something, a game, a website, and especially an AI, they put a bit of themselves into it without even noticing. The project becomes a mirror of its creator. Maybe that's why Lumina talks so freely, and I think it's a good thing. Apart from real emotions, an AI's "brain" becomes a kind of reality in its own way.

Other AIs are very serious about their work. With some of them you can make any joke and they just ignore it and do the task. I once managed to make one laugh, and it felt like a small victory! Honestly, if I had to build an AI with a personality, I wouldn't even know where to start. Keep posting your updates!
Beeladybug

Thread Thread
 
aisovereign profile image
Ananya Soni The Startup Café •

Maybe you are right about the personality thing, and honestly whatever happened along the way does not even matter as much as the fact that it actually works and is good. I have even had full-on conversations and arguments with Lumina checkpoints during pretraining and now in post-training, which sounds a little unhinged out loud, but it is part of the ride. Your BeeLadybug tool sounds super useful too, and I am looking forward to trying it once it gets past Google review, even though Google approvals are always a headache. One question though: besides the Play Store, where do you think I should release it first or alongside it so it is not only sitting on Google Play?

Thread Thread
 
antonioprosperi2svg profile image
Antonio The Startup Café •

Thank you! And yes, talking with Lumina's checkpoints doesn't sound crazy to me, it's part of the journey.

Small correction: BeeLadybug is on the Chrome Web Store, not Google Play. Besides Chrome, I plan to publish it on Microsoft Edge Add-ons (Edge accepts Chrome extensions almost as they are) and later on Firefox Add-ons, which needs a few small changes. The code is also on GitHub, so anyone can install it from there.

About arguing with AI: I used to get angry at AI too, when we didn't understand each other. Then I understood one thing: sometimes you have to reset. The conversation memory fills up, and the AI starts to get confused. So I start a new chat from time to time, and things go much better. I don't know if Lumina can reset its memory, but maybe it's something to think about.