cd /news/ai-research/a-signal-engine-with-no-training-no-… · home topics ai-research article
[ARTICLE · art-51202] src=github.com ↗ pub= topic=ai-research verified=true sentiment=· neutral

A signal engine with no training, no tokens, no neural net URL

James Jardine released AOI Shell v1.1, a token-free signal analysis engine using 96D octonion collapse and topological methods without training or neural networks, under AGPLv3. The tool runs offline and demonstrates applications in MRI quality control, LIGO glitch classification, and battery capacity tracking, with honest reporting of limitations.

read2 min views1 publishedJul 8, 2026
A signal engine with no training, no tokens, no neural net URL
Image: source

A topological "collapse" engine that reads structure in signals with no training, no tokens, and no neural network. It runs offline on a laptop — if you don't have a quantum sampler it falls back to local simulated annealing automatically, no account or key required. Clone it, run it, get the numbers below.

Built over five months by one person and an AI, and given to the world under copyleft. Institutions optimize for what they can measure; a lot of value falls through the cracks they can't see. This is a small tool for leveling that field.

These demos are training-free topological analyses (persistent homology). No model is fit to any data. The same mathematical tool is applied across three unrelated domains. We show you where it fails, too — that's the point.

As MRI k-space is undersampled, image loop-topology (Betti-1) degrades monotonically — no reference image, no training. Replicated across 6 brains (OpenNeuro ds000102):

acceleration 2.9×
Betti-1 (mean) 112 94 71 56 34 23

python demos/mri_qc.py

On the two most look-alike glitch classes, Blip vs Koi Fish, where peak amplitude barely separates them (AUC 0.24), topology does (AUC 0.84). Honest limit: on the easy, already-separable pairs a trivial spectral-bandwidth feature beats topology. This is a narrow result, not a universal one.

python demos/ligo_glitches.py

Topology tracks capacity fade strongly (r ≈ −0.7 to −0.9 across cells). Honest limit: it does not forecast future capacity beyond what you'd get from current capacity + recent trend (partial r ≈ 0.04). Descriptive positive, predictive null.

python demos/battery_predict.py

engine/

is the full 96D octonion-collapse organism: entropy gating, Fano/octonion projection, Jordan-Shadow decomposition, 33 transposable-element families, Monster-moonshine grading, and chain-complex homology. Token-free, no LLM, runs offline.

import numpy as np
from aoi_collapse_96d_dwave import aoi_collapse_96d_dwave

out = aoi_collapse_96d_dwave(np.random.normal(size=96))
print(out["betti"], out["intent"], out["chaos"], out["backend"])
pip install -r requirements.txt
bash fetch_data.sh   # downloads the public datasets into ./data/ (MRI, LIGO, battery)

Then run any demo from the repo root, e.g. python demos/mri_qc.py

.

AGPLv3 (see LICENSE

). You may use, modify, and redistribute freely; if you run a modified version as a network service, you must release your source. Every copy carries attribution (see NOTICE

). Commercial licensing (to use it without AGPL obligations): ** james@lattice24.com**.

James Jardine, AOI Shell v1.1: Token-Free Domain-Agnostic Signal Organism via 96D Octonion Collapse. DOI 10.5281/zenodo.20200607

James Jardine — Lattice24 / VoodooAOI · james@lattice24.com Built in collaboration with Claude (Anthropic).

── more in #ai-research 4 stories · sorted by recency
── more on @james jardine 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/a-signal-engine-with…] indexed:0 read:2min 2026-07-08 ·