DJC
Dennis J. Carroll
Hello, I'm

Dennis J. Carroll

Interactive ML Tools • Bayesian Analytics • Creative Fiction

I build interactive tools that make complex ideas intuitive. Over the past 5 years, I've created 26+ standalone web applications spanning neural network visualizations, Bayesian analytics dashboards, real-time ML training environments, and mathematical exploration tools — all designed to run in the browser.

My current focus is at the intersection of deep learning and interpretability: understanding not just what neural networks learn, but how and why they learn it. Recent projects include an Agent Trace Viewer for SWE-agent interaction analysis and a Mechanistic Interpretability visualizer for transformer architectures.

I'm also the author of three original fictional universes — including Crack in the Veil, a post-humanity sci-fi saga, and A Chronicle of Lyos, a fantasy world where dead gods' bloodlines still remember.

By the Numbers

26
Interactive Apps in the Browser
5+
Years Building
3
Original Fictional Universes
317
Tests Passing in GLASSPORT

Skills & Technologies

Programming Languages

Python
JavaScript
TypeScript
SQL
R

Machine Learning & Deep Learning

TensorFlow
PyTorch
TensorFlow.js
Scikit-learn
Bayesian Inference
Mech. Interpretability

Data Science & Analytics

NumPy
Pandas
PyMC
Streamlit
Jupyter
Statistical Modeling

Web Development

React
Gatsby
Tailwind CSS
Three.js
Framer Motion
HTML5 / CSS3

Tools & Infrastructure

Git / GitHub
Docker
AWS / Cloud
Linux

Experience

Independent Software Developer & Data Scientist

Self-Directed
Current
Remote 2019Present
  • Built 26+ browser-based interactive applications — neural network visualizers, Bayesian analytics dashboards, mechanistic interpretability tools, generative audio/visual systems — all running in the browser with no install required.
  • Designed and implemented ML pipelines in Python using TensorFlow, PyTorch, Scikit-learn, and PyMC; deployed interactive frontends with React, Gatsby, Three.js, and TensorFlow.js.
  • Authored research-level interpretability tooling for transformer model analysis, including an attention-head visualizer and a structured annotation system for MLP blocks.
  • Wrote and self-published three original fictional universes — a post-humanity sci-fi saga, a space opera, and a secondary-world fantasy — developed in parallel with technical work.
PythonTensorFlowPyTorchReactThree.jsBayesian InferenceGatsby

Doorman / Building Security

Residential Building
Current
New York, NY 2018Present
  • Primary point of contact and access control for a high-occupancy residential building — trusted with building security, resident communication, and emergency response.
  • Managed relationships with residents, management, vendors, and emergency services; developed strong judgment under pressure in a high-visibility, low-error-tolerance role.
  • Applied communications degree background daily: de-escalation, conflict resolution, clear documentation, and coordination with building staff.
Security ProtocolsCommunicationConflict ResolutionEmergency Response

Construction & Site Work

Various Projects
New York Area 20152019
  • Performed skilled labor across multiple construction sites — developed strong work ethic, spatial reasoning, and understanding of project sequencing under tight deadlines.
  • Collaborated with crews on-site, reading plans and coordinating tasks; skills in logistics and structured problem-solving carried directly into software project work.
Project CoordinationSite SafetyTeam Collaboration

My Journey

How I build

Most of my tools run entirely in the browser — no backend, no install. If an idea about neural networks or Bayesian inference can't survive being made interactive, I don't trust that I understand it yet. Building the visualization is how I find out.

Why the fiction

The worldbuilding is the same skill pointed elsewhere: take a system with rules, push it until it breaks, write down what happens. Three universes so far, each one built around a single broken premise.

Current Work

GLASSPORTIN THE WORKS

Wire-level observability and enforcement for MCP servers: passive stdio tap, behavioral detectors, data-exfiltration scanning, and SARIF export to the GitHub Security tab. Python 3.10+, zero runtime dependencies, 317 tests.

View on GitHub →

Mechanistic Interpretability

Understanding how transformer attention heads and MLP blocks represent concepts

See experiment →

Graph Neural Networks

Extending neural architectures to non-Euclidean graph-structured data

Reinforcement Learning Theory

Connecting RL optimization to dynamical systems and chaos theory

See experiment →

About This Website

React
+
Gatsby
+
Tailwind
+
Framer

Gatsby and React render the shell; the experiments themselves are standalone static apps, so each one loads without pulling in the rest of the site.

Open to collaboration and interesting projects
© 2026 Dennis J. Carroll. All rights reserved.