Paul.Whitten.dev

Current work

Deterministic-Dimension Reduction for Reliable LLM Agents: A Design-Science Study of Spatial Construction and Edge Deployment

Paul Whitten, Li-Jen Chen, Sharath Baddam - accepted to HICSS 60

This work explores deterministic-dimension reduction techniques for enhancing the reliability of LLM agents in spatial construction and edge deployment scenarios.

Local Preprint

2.5-D Decomposition for LLM-Based Spatial Construction

Paul Whitten, Li-Jen Chen, Sharath Baddam - accepted to NAECON 2026

A neuro-symbolic pipeline where the LLM plans in the 2D horizontal plane while a deterministic executor computes vertical placement, removing a whole class of errors by design.

Read on IEEE Xplore arXiv Preprint

Explainability Methods for Hardware Trojan Detection: A Systematic Comparison

Paul Whitten, Francis Wolff, Chris Papachristou - accepted to the Journal of Electronic Testing (Springer Nature)

A systematic comparison of three categories of explainability for gate-level hardware trojan detection on the Trust-Hub benchmark: domain-aware property-based analysis, model-agnostic case-based reasoning, and model-agnostic feature attribution (LIME, SHAP, gradient).

Read full text

AgentX - AgentBeats

Team AgentWhetters - Berkeley RDI

1st place (tied), Gaming Agents track (Build What I Mean) - repo
1st place, Coding Agents track (SWE-bench Pro) - repo
1st place, Sprint 4 general-purpose agent - repo
2nd place, Cybersecurity Agent (CyberGym)

Autonomous coding agent

Open source - TypeScript

A multi-agent autonomous coding framework on the GitHub Copilot SDK. Agents coordinate via a mailbox protocol and run JSON workflow state machines with deterministic quality gates.