Independent research & engineering

Casey
Zak

AI systems, computational R&D, and the occasional question that refuses to stay theoretical.

Alongside my professional engineering work, I maintain a small independent R&D practice: a place to build experimental AI applications, computational tools, and systems that begin as questions rather than product requirements.

Independent work

A private laboratory for systems that are worth understanding.

01 / SYSTEMS

AI & Systems

Agentic applications, inference systems, software infrastructure, and computational tooling — built to understand how capable systems behave in practice.

02 / STRUCTURE

Mathematical & Computational Research

Exploratory work in dynamical systems, geometry, simulation, and mathematical models, often used to make abstract structure concrete.

03 / EVALUATION

Security & Evaluation

Adversarial evaluation, failure-mode analysis, and defensive testing of AI and software systems in authorized environments.

Research study / Hermitia

Light, viewed twice.

One prescribed mode, two representations. Read its fields and the Hermitia candidate’s signed loop phases in the same cell space, then follow their phase progression.

Shared cell space

X–Z slice · select a cell to inspect its loop phase.

Signed loop phases and canonical mode components share one lattice slice.
Loop phase φ · same scale in every view
−π/20+π/2 rad

Phase progression

Earlier at the top, current state at the bottom. Both strips use the same cell positions and prescribed history; scrubbing updates the whole pattern.

X–Z loop phaseMarked cell now 0.000 rad
Space–time history of the X–Z loop phase.
Y–Z loop phaseMarked cell now 0.000 rad
Space–time history of the Y–Z loop phase.

Canonical photon

Electric · xMagnetic · x

The slice shows the in-plane components of a transverse field mode. These are mode amplitudes, not a particle trajectory or a single photon’s mean electric field.

Hermitia photon candidate

Signed loop phase

Teal and amber separate opposite phases; the small hands show their angles. Changing circular handedness reverses the Y–Z phase pattern. Rephasing individual links leaves both loop patterns fixed.

X–Z loop phase 0.000 radDefect 0.000

Paused at the center. Read the signed phases face-on, or play to follow their progression.

One supplied mode, two representations. The loop phases are computed from the same potential as the canonical field mode, on an imposed π-flux cell background. The animation illustrates a relational encoding; it does not establish a separate Hermitia propagation law or a physical difference from conventional light.

What the light is measuring

The opened cell's anticommuting pairings give a background square holonomy of −1. With an added loop phase φ, W = −exp(iφ). The face color shows signed φ on a fixed −π/2 to +π/2 scale; the supplied profile stays within that range. Small phase hands point right at φ = 0 and turn counterclockwise for positive φ. Loop orientation is +x, +z, −x, −z in the X–Z slice and +y, +z, −y, −z in the Y–Z slice. The separate defect readout is |1 + W|² = 2(1 − cos φ); it assigns the same value to opposite phases. Neither quantity is an energy or a detection probability. The extended lattice also has a baseline defect from straight two-step paths.

Both layers sample the same transverse potential profile. The canonical layer draws its electric and magnetic mode shapes; the Hermitia layer assigns its transverse samples to link phases. The visible lattice is a cropped window, and its spacing is a display choice. The supplied profile varies only along z, so parallel slices repeat the same phase pattern. History strips show the preceding quarter-pass and restart with each playback pass. Rephasing adds a difference of endpoint phases to each link, leaving closed-loop products unchanged.

Circular polarization supplies the handedness in this example. The profile has no transverse azimuthal phase winding, so this demonstration does not include an optical vortex carrying orbital angular momentum. Conventional electromagnetism already supports spin and orbital angular momentum; a distinct Hermitia prediction would require its own derived dynamics and a matched comparison. Spin and orbital angular momentum in electromagnetism.

Loop variables are also standard in lattice gauge theory. Hermitia's research question concerns their origin in relational closure and the cell's pairings. A later study can add the cell–mirror stitch and examine its spectral gap.

On the canonical side, a single photon occupies a field mode; these ribbons visualize that mode rather than a nonzero mean field. Photon wave packets and field modes.

Operating scope

Serious experimentation, deliberately bounded.

The practice is intentionally small. It exists for independent technical work outside my primary professional role. Security testing is limited to systems I own, control, or am explicitly authorized to assess; the public surface stays intentionally sparse.

FocusAI · systems · computation
Security postureDefensive / authorized evaluation
ModeResearch · build · measure
Public infrastructure detailIntentionally omitted
Field note / 001

The background is doing real math.

It is not a stock animation. Each material point is a normalized two-component spinor. A spatially varying Hermitian Hamiltonian evolves the field exactly and unitarily; the visible geometry is built from its Bloch bilinear. The trailing filaments are nearby moments of the same conserved flow.

Dynamicsi ∂t ψ(s,t) = H(s) ψ(s,t)
Hermitian generatorH(s) = h(s) · σ = H(s)†
Observablen(s,t) = ψ† σ ψ  ∈ S²
This site is served as static assets: no analytics, trackers, forms, external fonts, third-party scripts, or network calls.