Runtime observability
Token-level telemetry that maps how language model trajectories move, branch, and respond to perturbation.
// Josh Malone — AI Systems Builder
Building AI systems.
I build research instruments and operator-grade tools for AI — runtime instrumentation, perturbation testing, and security visibility for the questions that don't have clean answers yet.
Deterministic comparisons
Branchpoint snapshots and config hashes so baseline-vs-intervention runs actually mean something.
Shipping discipline
Small, usable tools with a bias toward real workflows. Built fast. Tuned in the open.
// Featured.Project
Observer
v0.2.0 · Phase 2 — mapping
Observer is a runtime research lab for language-model trajectory mapping: how generation moves, branches, persists, and responds to perturbation at token time.
- Hysteresis protocol — does perturbation memory persist after the perturbation is removed?
- Observability runner — token-level divergence, spectral diagnostics, and SVD signatures
- Intervention engine — deterministic SeedCache branchpoints for honest baseline-vs-intervention comparisons
- Adaptive controller — closed-loop research mode (archived after v1 thesis was falsified in v2)
Divergence signal
VAR(1) · 64d
24.7k
Tokens mapped
Δ +1.2k
97.6%
SeedCache match
Δ +0.3%
// Selected.Work
Tools with a practical edge.
Observer
Runtime instrumentation for trajectory mapping in language models — perturbation, branchpoint, and control experiments at token time.
Open repoClawGuard
High-performance C++ system monitor and security watchdog — port intrusion detection, skill attribution, trend analysis, and proactive alerts.
Open repoGCT Multiplicity Lab
Research lab for GCT multiplicity-gap experiments with stretch gating, HWV/orbit triage, and adversarial closure probes.
Open repo
// About
Builder and researcher,
chasing the open questions in AI.
I'm an Air Force veteran based in Huntsville, Alabama. Self-taught, product-minded, and drawn to hard problems where reliability and clarity matter.
The through-line in my work is simple: build AI systems that can be inspected, tested, and improved by real operators.
USAF · Veteran
Self-taught
Huntsville · AL
Operator-first
// Establish.Connection
Build with me.
If you're working on AI reliability, evaluation, security, or operator tools — I'd be glad to talk. No pitch deck required.