Yeah, I know. My last post here was on April 27th.
If you are wondering what I have been doing for the last few months, no, I wasn't just staring at my wallpaper. I actually stopped posting because my last few posts got exactly zero engagement. Literally zero. But I figured it was time to break the silence because the stuff we have been building in the dark is finally ready for the light.
I am not going to turn this into a 10,000-word essay, so consider this the highlight reel of what I (and my company, AGI Systems Directorate) have been up to.
1. AGI Gauntlet & AGI Bench 2026
Have you ever wondered how we actually measure if an AI agent is getting smarter, or if it is just getting better at memorizing benchmark exams?
We were tired of the standard testing, so we built AGI Gauntlet. It is a Python library for AI and AGI development, and a strict benchmark for AI agents. You can pip install it right now. We also launched AGI Bench 2026, the web UI version comparing our projects against commercial models like Google Gemini.
Oh, and there is an Open Contributor Challenge running for the Python library right now. If you want the details on how to jump in, let me know in the comments.
2. Portfolios, Hugging Face Spaces, and Docker
Before we got completely derailed by our biggest project, I realized developers need better ways to show off their work. So I built a developer portfolio template. You can find it on my GitHub (the subdomain name is weird, you will see it when you find it).
We also pushed the official Docker packages to the GHCR registry and released Hugging Face Spaces for our previous models, like Velocity Nova Prime and A.R.N.A.V Omni.
3. Stop Slicing Your Text Like Salami
Are you still blindly splitting texts by character count for your RAG pipelines? Be honest.
I wrote a piece on HackerNoon about why that is a terrible idea and how to handle semantic chunking properly. If you are a developer looking for a tutorial on how to actually maintain context when chunking data, go check out my HackerNoon profile. It is a game changer for RAG setups.
4. Project Lumina: The Biological LLM
At AGI Systems Directorate, our initial goal was always to create artificial general intelligence. Everything else was just a warm-up.
Right now, we are working on our biggest model yet: Lumina.
Lumina is not just a large language model. It is an autonomous AI agent built entirely from scratch. No existing frameworks. No existing APIs. It is a hybrid architecture that combines standard transformer elements with actual biological mechanisms. We are talking frontal lobe fluid intelligence, synaptic plasticity, the free energy principle, and global workspace theory.
I am not going to list every single mechanism here because nobody wants to read official documentation in a dev post. But here is the best part: it is going to be fully open source and free to use. Both the API and the chat interface. I am completely tired of the current API costing structures in the AI space, so we are just removing the cost barrier entirely.
Lumina weights will be out soon on Hugging Face.
What's Next?
Like I said, this is just the surface level. I don't want this post to be a monologue. If you want a deep-dive tutorial or more technical details about anything I just mentioned, whether you are a beginner or an advanced dev, drop a comment below. I actually read them.
And if you found any of this even slightly interesting, hit the unicorn button. Let's see if we can break my zero-engagement streak.
Top comments (0)