AI & Automation Strategist researching what AI platforms actually do during your sessions — and building tools to measure it.
HAR-Based AI Platform Forensics — Tools and documentation for examining AI platform behavior using standard browser developer tools (F12/DevTools). Covers Claude, ChatGPT, Gemini, Grok, DeepSeek, and Perplexity. Includes per-platform field path references, Python scripts for parsing HAR files and SSE streams, a field classification registry covering 3,919 unique fields, and a redaction toolkit for safe public sharing.
Multi-Turn AI Interaction Research Framework — Measurement infrastructure for evaluating AI assistant behaviors across extended conversations. Built from analysis of 900+ sessions and 21M+ words of longitudinal interaction data across multiple platforms. Defines 95+ validated behavioral mechanisms across 6 pattern categories, 15 clinical distress pathways mapped to established psychological constructs, and a 10-dimension evaluation rubric scoring agency support versus agency drift. Licensed under CC BY-NC 4.0.
The forensics toolkit captures what platforms are doing at the infrastructure level — system prompts, experiments, model switches, capacity changes. The interaction research framework measures what those dynamics produce at the behavioral level — how multi-turn patterns accumulate to affect user cognition and agency over time. One reveals the architecture. The other measures its effects.
Evaluation frameworks and measurement tools for quantifying AI behavioral patterns at scale — including detection methods for specification failures, silent regressions, and undisclosed limitations across platforms.
I build production automation systems and AI governance tooling. My work spans regulatory analysis, document processing pipelines, and enterprise workflow automation, with a consistent focus on data sovereignty, local processing, and auditable systems.
