Building the Tesseract: What Happens When an Archive Learns to Read Itself? Part 1

6/10/26

Over the past week I’ve been working on something that started as a technical experiment and turned into one of the more interesting investigations I’ve done in years. The original idea was simple enough. I wanted to see what would happen if I gave AI access to my archive and allowed it to analyze twenty years of artwork, writing, teaching, experimentation, and documentation. Not a few selected images. Not a curated portfolio. The archive itself.

At the moment that archive consists of more than 1,000 published blog posts and essays, over 9,000 images in my WordPress media library, and work spanning roughly 2006 through 2026. What makes this even more interesting is that the public archive only represents part of the story. Sitting outside of WordPress are thousands of additional photographs, drawings, paintings, animations, source files, scans, installation images, videos, and documents spread across hard drives, cloud storage, old computers, and various digital graveyards accumulated over the last twenty-five years. The first version of the project became NET-ART OS, an archive intelligence system designed to ingest, organize, search, and analyze large collections of creative work. Initially I thought I was building a better archive search engine. Something that could identify relationships between artworks, surface forgotten projects, and help me navigate decades of material more efficiently. That alone would have been useful.

One detail that’s important to mention is that none of this happened inside a polished software platform. There was no development team, no research lab, and no enterprise infrastructure behind it. The entire project began on my MacBook Pro after installing Claude Code and pointing it at my own archive.

The workflow itself became part of the experiment.

Throughout the process I moved continuously between ChatGPT 5 and Claude Code. ChatGPT acted as a strategic collaborator, helping frame questions, challenge assumptions, identify blind spots, and suggest new directions. Claude Code operated inside the terminal as a builder, researcher, analyst, and implementation partner. Ideas often originated in one environment and were tested in the other. Discoveries made by Claude were challenged through conversations with ChatGPT. Questions raised by ChatGPT became new experiments executed by Claude. The process became less about using AI tools individually and more about orchestrating a conversation between multiple forms of intelligence.

The archive itself was powered by WordPress. Using the WordPress REST API, thousands of posts, images, metadata records, categories, tags, and media assets were ingested into a local archive intelligence system. Claude Code helped build NET-ART OS, transforming that material into a searchable and analyzable corpus. The system relied on Python, SQLite, embeddings, metadata extraction, clustering, statistical analysis, and archive retrieval pipelines running locally through the terminal. What I find most interesting is how accessible this process actually was. The entire experiment was conducted using a personal archive, a MacBook Pro, Claude Code, ChatGPT, WordPress, Python, and open-source tooling. No custom hardware. No venture funding. No specialized research environment. Just twenty years of accumulated work meeting a generation of tools that did not exist when most of that work was originally created.

What happened next surprised me.

The archive started revealing patterns that I hadn’t consciously recognized myself. Certain themes kept returning. Certain questions seemed to persist regardless of medium. Ideas would appear in one form, disappear for years, and then reappear through an entirely different technology. A drawing from one decade would unexpectedly connect to a GIF from another. A sculpture would echo a blog post written years later. The archive wasn’t behaving like a collection of files. It was behaving more like a system.

Somewhere along the way the project became what Claude and I started calling the Tesseract, borrowing inspiration from Interstellar, one of my favorite films. The idea was less about science fiction and more about navigation. What happens when an archive stops being chronological and becomes relational? What happens when twenty years of work can be explored through recurring questions, visual similarities, conceptual relationships, and unexpected connections rather than folders and dates? As the project evolved we expanded beyond text and began analyzing images. This became the Visual Tesseract. More than a thousand images were embedded and clustered. Visual motifs started appearing across years. Certain color relationships kept resurfacing. Similar compositional structures emerged between works that had never been intentionally linked. Some of the visual findings appeared to support discoveries that were already emerging from the textual analysis. For a brief moment it felt like the archive was beginning to describe itself.

This is also where things became dangerous.

AI is exceptionally good at generating convincing stories, and convincing stories are not the same thing as evidence. Some of the findings felt profound. Others felt suspiciously perfect. At that point the project shifted from discovery to skepticism. Instead of asking what new theories we could build, we started asking how many of our favorite ideas would survive being attacked.

What followed was probably the most valuable part of the entire process.

An adversarial audit was conducted across dozens of project documents. Every major claim was challenged. Contradictions were identified. Definitions were tested. Assumptions were dragged into the open. The project was effectively forced to argue with itself. The audit separated the work into three layers: methodology, theory, and philosophy. That distinction turned out to be critical.

One of the most interesting findings didn’t survive.

For several days we believed that experimentation represented the deepest invariant in the archive. The evidence seemed compelling. The word appeared across all nineteen years of published content. It looked like a throughline running across the entire body of work. Then we built a permutation-based null model and tested it.

The result was immediate and humbling..

The finding collapsed..

What looked like a profound structural truth turned out to be statistically indistinguishable from a common high-frequency word appearing throughout a large corpus. In short, the archive had fooled us. Oddly enough, that failure increased my confidence in the methodology. The system had just disproved one of its own favorite conclusions. That’s exactly what it should do. If every result confirms the theory, you’re no longer doing research. You’re doing mythology.

The more recent tests have been far more interesting. Some findings disappeared under scrutiny while others became stronger. The archive’s accessibility and deafness-related themes emerged as genuine long-term signals. The rise of AI and agent-based systems appeared as a measurable historical event within the archive itself. Even more interesting, thematic structures from earlier periods of the archive demonstrated an ability to predict aspects of later periods better than chance. In other words, some parts of the archive genuinely contain information about where the archive is likely to go next.

At this point I no longer think of NET-ART OS as a search engine, a product, or even an archive project. The best way I can describe it is as an instrument. A telescope pointed inward. Something capable of revealing structures that are difficult to perceive manually across decades of creative work.

There is still a tremendous amount left to do. The Visual Tesseract is only partially built. The larger unpublished archive remains largely untouched. The spatial computing, XR, VR, and mixed reality components exist mostly as ideas and prototypes. The methodology itself has only been tested against a small number of archives. There are more questions than answers. What surprised me most about this process is that the most valuable moments weren’t the ones where the system confirmed something I already believed. The most valuable moments were the ones where it contradicted me, challenged assumptions, or revealed relationships I had never noticed. Those moments are rare. They are also the reason I’m continuing.

The work already existed. The archive already existed. The questions already existed. What changed was the arrival of systems capable of reading that archive at scale. In many ways, the archive was simply waiting for the technology to catch up.

For now, I’m taking a short break from building and documenting what happened. The archive is still there. The questions are still there. The Tesseract is still there. The experiment continues.

Want more?

Relevant posts and follow ups: