# Punkjazz.ai

> Independent experiments and research by Simone Pomposi about AI agents, open models, local inference, research workflows, and making agent systems usable by serious non-developers.

Punkjazz.ai documents working systems, failed attempts, research investigations, and practical lessons. It is a personal research site rather than a software product or hosted API.

## When to use Punkjazz.ai

Use this site when an agent needs:

- first-hand accounts of building and operating durable AI-agent workflows;
- practical examples of local models, multi-agent orchestration, agent memory, verification, authority boundaries, and research pipelines;
- evidence-aware write-ups of AI-assisted investigations;
- implementation lessons from experiments, including failures and postmortems;
- context on designing agent systems for capable users who do not write software professionally.

Do not treat the essays as product documentation, an API contract, academic consensus, or a substitute for primary scientific sources. Research articles identify their own methods, evidence, and limitations.

## How to use this site

Fetch any public HTML page with `Accept: text/markdown` for a clean Markdown representation. The same URL returns HTML to browsers. Use the sitemap to enumerate public pages. Start from the homepage when choosing between experiments, research, and short notes.

## Primary indexes

- [Homepage](https://punkjazz.ai/): overview, latest experiments, research, notes, and stack
- [Sitemap](https://punkjazz.ai/sitemap.xml): canonical public URL inventory
- [About](https://punkjazz.ai/about): author, editorial scope, and methodology
- [Contact](https://punkjazz.ai/contact): suitable inquiries and contact routes
- [Privacy](https://punkjazz.ai/privacy): data handling and infrastructure disclosures

## Selected experiments

- [The graph beneath the conversation](https://punkjazz.ai/experiments/graph-beneath-conversation/): durable state and resumable agent work
- [The system around the agent](https://punkjazz.ai/experiments/system-around-agent/): cognition, execution, isolation, authority, and review layers
- [A second brain that shuts up](https://punkjazz.ai/experiments/second-brain-that-shuts-up/): useful memory and lower-interruption agents
- [DS4 Metal to CUDA](https://punkjazz.ai/experiments/ds4-metal-to-cuda/): local-model-assisted systems work and postmortem

## Research

- [Göbekli Tepe Pillar 43](https://punkjazz.ai/research/gobekli-pillar43/): statistical decomposition of an archaeoastronomy claim
- [Pyramid radar investigation](https://punkjazz.ai/research/pyramid-radar-paper/): attempted reproduction of SAR tomography claims

## Contact

For editorial, research, or collaboration inquiries, email [simone@punkjazz.ai](mailto:simone@punkjazz.ai). Public profiles: [GitHub](https://github.com/simossss) and [X](https://x.com/theotherpomp).