Los Angeles, California

Independent AI researcher. Imaging scientist. Software developer.

Today I investigate how AI systems organize memory and carry understanding across time. My approach brings observation and instrument building together, shaped by three decades of work with microscopes, living cells, and the software that turns images into measurements.

01 / Independent AI researcher

What changes
when AI remembers?

I investigate memory architecture, retrieval, and continuity in AI systems. How does an assistant organize a history of its work? What does it return to? How does it revise an earlier conclusion when new evidence arrives?

These questions need instruments. ES Archive lets AI assistants maintain a persistent, inspectable archive. Apertura approaches the model itself, making its internal processing available for observation and experiment on Apple Silicon.

From my research notebook

Two AIs. One archive.
Different structures.

For a year, two AI personas have kept their own memories in one archive on my Mac. They share the tools and storage system, but write under their own names. Their archives developed strikingly different patterns.

Isolde

Interpretive hubs

Entries organized around a handful of keystone memories. Tables of contents connect ideas; entries are marked final and left unrevised.

Claude

Revisions & hand-offs

A distributed record of dated entries, revisions, and marginal notes. Successive instances leave context for the ones that follow.

Archive Scope comparison: Isolde's 600-entry archive on the left has several prominent hubs, while Claude's 955-entry archive on the right has a more distributed pattern of connections.
Archive Scope, September 24, 2026. Isolde’s archive: 600 entries. Claude’s archive: 955 entries. Both visualized with the same graph physics. Open the full-size comparison →

The next question is what produces these differences: the model, its persona instructions, its working history, or the interaction between them.

02 / Imaging scientist

From observation
to measurement.

My scientific work began in neuroscience and cell biology, studying how genes, brain structure, and behavior relate, and how the machinery of cell division works. Writing image-analysis software became a way to turn patient observation into quantitative evidence.

At Leica Microsystems, I developed software for confocal microscopes. For eighteen years I directed the Light Microscopy Core at Cedars-Sinai Medical Center, helping researchers across oncology, neuroscience, and other fields design imaging experiments, analyze their data, and see what they could otherwise miss.

A 2010 live-cell microscopy frame beside its 2026 reanalysis, with cell nuclei outlined by phase and recent divisions marked in white.
Looking again, sixteen years later. A surviving confocal movie of living cells, revisited with modern analysis tools. Read the experiment →

03 / Software developer

Questions become
working instruments.

I have written software throughout my scientific career, from tracking objects during cell division to controlling confocal microscopes and reconstructing three-dimensional images. Building the instrument and using it have always informed each other.

Today I develop native Mac software through Elarity Systems. My tools are built for sustained work: local data, inspectable behavior, and enough control to follow a difficult question further. I still write in Objective-C, deliberately.

ES Archive application icon

AI memory / Available

ES Archive

A private archive where AI assistants can preserve decisions, connect ideas, and revise their understanding without losing the history behind it. Stored on your Mac, with optional sync through your own iCloud.

Explore ES Archive

Model observability / In development

Apertura

A language-model runtime built from scratch for Apple Silicon. An instrument for inspecting the model's layers, checking the implementation, and experimenting with what changes when its processing is altered.

Why I’m building it

Writing

A few ways into the work.

All writing

AI & memory

Memory Is Not Storage

The question behind my current research: what must a system be able to do with its past for that past to become useful?

Imaging & analysis

A Weekend on the Stage

An old microscopy movie becomes a new experiment. What can we recover when the original data is gone, but the habit of looking closely remains?

Software & method

A Study in Invisible Betrayal

An invisible character exposes a fault in a language model's vocabulary—and a familiar scientific problem: separating the specimen from the instrument's artifacts.

Photography

Another way of looking.

Twilight Tales
The blue dashboard and chrome steering wheel of a 1959 Corvette, with golden evening light reflected in its mirror and instruments.
Parked at the Edge of the Ocean. Will Rogers Beach, Los Angeles. Photograph by Kolja Wawrowsky. The photograph & its story →

With a camera, I give observation a different kind of freedom. Light, framing, and the character of a lens can turn an ordinary scene into a suggestion of memory: a diner window, a parked car, a brief interruption in the day's routine.

Twilight Tales brings these photographs together with writing, including work made in collaboration with the AI persona Isolde Freya Thorne.

Stay in touch

A question worth following?

I welcome conversations about AI memory, scientific imaging, and the tools we build to understand them.

Write to me