Project summary

I custom built an AI at home. Her name is Ava. She's modeled after the human brain. Real spiking neurons. Memory that changes and adapts while she reads. A cerebellum that keeps learning while you talk to her. Dopamine, mood, a thalamus, all of it. She's written completely in Rust, no Python anywhere, and I trained her on one GPU, a 4070. She's open source and free, anybody can download her and talk to her: https://github.com/nochallenge/ava

What are this project's goals? How will you achieve them?

I believe the fastest way to achieve ASI is to model the code after the human brain. Ava is very small right now, 27 million weights. I want to find out if her design keeps up with a transformer when she gets bigger. So the goal is one test. Train Ava and a normal transformer at 100 million weights, same data, same budget, and compare them. I write the rules down before the run so the result is honest.

How will you know it worked?

I get two numbers side by side, Ava and the transformer at 100M weights. If Ava is close or ahead, her design is worth scaling further. If she's behind, I publish that too. On my 4070 she was ahead of the transformer at 1M and 3.3M weights and a little behind at 17.8M, and that's as far as my card can take her.

How will this funding be used?

All of it goes to renting GPUs. My 4070 has 12 GB and this run doesn't fit on it. I need a 24 GB or bigger card for a few hundred hours.

Who is on your team? What's your track record on similar projects?

It's just me. I built Ava and released four versions of her between October 24 and November 11, 2026. Ava 0.4 comes with a chat app where you can watch her brain light up while she thinks, and that's not an animation, it's her code mapped to the parts of the brain it's modeled after. I also built and run Noozra (noozra.com) by myself, a news, community and jobs platform that's been live since September 2026.

What are the most likely causes and outcomes if this project fails?

The most likely way it fails is Ava loses to the transformer at 100M weights. Then I know her design has to change before anybody spends real money on it, and everybody else gets to see that result for free. The other risk is my code needing work to run on a bigger GPU, so I'll do short test runs first before the long ones.

How much money have you raised in the last 12 months, and from where?

$0. Ava has been paid for out of my own pocket. I applied for GPU credits from CloudRift, Lambda and Hanzo and haven't heard back yet.