# Misul Computing

> Misul Computing is an independent AI research lab in Sofia, Bulgaria. The lab builds machine intelligence from the principles of the brain: how it perceives, predicts, learns, and gives rise to mind.

Misul Computing is independent and research-led. It designs new architectures from first principles rather than by scaling existing methods.

## Key facts

- Name: Misul Computing
- Type: Independent AI research lab
- Location: Sofia, Bulgaria
- Founder: Deyan Todorov
- Contact: deyan@misul.org
- GitHub: https://github.com/Misul-Computing

## What the lab is working toward

Machine intelligence built from the principles of the brain. Systems that:

- perceive and predict the way the cortex does
- learn continually from a stream of experience
- reason and plan over what they learn

## Public work

Monodratic is the first public piece. It is a technical report on a stateless sparse attention mixer. After rotary position embedding, product routing writes keys into bounded causal posting lists and selects a fixed number of remote blocks. Attention over that set is exact causal softmax.

Grounded results from the technical report (10 July 2026):

- Associative-recall accuracy: 0.9909 mean, 0.9805 minimum, three seeds, routed width 2 of 5.
- Minimum margin over equal-width random routing: 0.4297.
- Minimum margin over local-only attention: 0.7070.
- Target-block hit rate: 0.9362 mean.
- Selected-mask oracle error: 1.91 x 10^-6 maximum.
- Overflow: 0.
- Packed timing exponent: 1.004 from 4,096 to 32,768 tokens on four CPU threads.
- Component tests: 429 passed.

Monodratic does not claim better language-model quality than other architectures. It does not claim a fused-kernel speedup. It does not claim cache reduction.

Paper: https://github.com/Misul-Computing/Monodratic/blob/main/output/pdf/monodratic_proof.pdf
HTML: https://misul.org/monodratic/
Code: https://github.com/Misul-Computing/Monodratic

## Pages

- [Home](https://misul.org/): The lab, its research direction, and public work.
- [Monodratic](https://misul.org/monodratic/): Technical report. Mixer, figures, evidence, limits, PDF, and code.