March 30, 2026
What Dead Code Taught Us About Building Tools for AI Agents
A benchmark of graph-enhanced vs baseline AI agents on dead-code detection across 14 production repos.
Building code intelligence systems
I build tools that help developers understand codebases. Currently focused on graph-based context retrieval, cross-file call resolution, and LLM-powered code analysis.
Demo
Progressive graph updates visualized in the Supermodel VS Code extension. We build a ground-truth architectural graph using our analysis algorithms, then watch filesystem changes to provide live updates to the architectural map of the program.
Projects
onehand ✋
Control your Mac from your phone with one hand. Your phone becomes a trackpad and dictation mic for a Mac across the room. LAN-only WebSocket bridge, PIN-paired, no cloud, no accounts.
Built while my newborn daughter slept on me after a feed. A sleeping baby pins you to the couch for up to two hours with exactly one hand free, and that hand has a phone in it. The Mac was across the room plugged into the TV. onehand closed the gap: a trackpad with tap, scroll, and drag gestures, plus continuous dictation that types into whatever app has focus. Designed, built, and debugged one-handed, using itself.
Python
JavaScript
macOS
Android
WebSockets
Clamshell Sentinel
macOS menu bar app that keeps a MacBook awake on clamshell close only while configured agent or terminal processes are running.
Swift
macOS
Developer Tools
Supermodel
Code intelligence platform providing graph-based context retrieval for AI coding assistants. Built call graphs, dependency graphs, and domain classification using tree-sitter parsing and Louvain community detection.
TypeScript
Neo4j/Cypher
MCP
Tree-sitter
Ghostie
CLI tool for managing AI context and memory across terminal sessions. Features knowledge graph capabilities with hashtags and wikilinks for persistent context.
JavaScript
CLI
npm
Age of Claude
Experimental project exploring AI agent interactions and emergent behaviors in simulated environments.
Python
Claude API
Writing
All posts →
Experience
Founding Engineer
Jan 2025 - PresentSupermodel
- Designed the LLM-driven domain classification shipped as the
/v1/graphs/domainpublic endpoint, with multi-provider routing across Anthropic, OpenRouter/Gemini, and Cerebras - Built the parse, dependency, and call graphs and the analytics behind four production endpoints (dead-code, impact, test-coverage, circular dependencies)
- Authored the dead-code benchmark: graph-enhanced agents vs. baseline Claude Opus 4.6 across 14 production repos — 94.1% avg F1 at 100% precision vs 52% baseline, 156× cheaper, 11× faster
Principal
Mar 2022 - PresentScale Factor LLC — Sylvester, Georgia
Software engineering research & development consultancy. Selected client engagements:
ACI Seeds — Data Automation
2024- Built a Python/Pandas pipeline to process gas chromatograph results for high-oleic peanut variety trials
- Reduced document processing from 4 hours to 15 minutes per 200-page report (16× improvement); scaled the lab from 2 reports/day to their full annual volume (~100 runs)
Everyday Futures LLC
Mar 2022 - Jan 2023Software engineering contract.
Technical Skills
Languages
TypeScript
Python
Java
Cypher
Go
Technologies
Neo4j
Tree-sitter
MCP
Docker
VS Code Extensions
AI/ML
Claude/Anthropic
OpenRouter
Structured Outputs
RLHF Training
Infrastructure
Azure
Terraform
CI/CD
Containers
Education
BS Mathematics
Indiana University
Real Analysis, Abstract Algebra, Topology, Probability Theory, Mathematical Modeling
AS Computer Science
Georgia State University
Data Structures, Algorithms, Discrete Mathematics
Speaking & Community
GDG Atlanta Founder
Speaker at DevNexus, Connect.js, CodeMash