🧠 Primary Focus Areas
1. LLM Behavior & Rubric Evaluation
- Developed detailed rubrics for assessing conversational reasoning and model behavior
- Identified and categorized LLM failure modes across:
- Instruction retention
- Inference memory
- Versioned editing consistency
- Self-coherence under adversarial input
- Co-developed
multi_challenge_faults_reference.md as a canonical logic fault matrix
- Actively contributed to Meta-Evaluator pipelines with roleplay, contradiction, and mood logic probes
2. Prompt Engineering Systems
- Co-developed the Dreadnaught system for diffusion-based image prompting:
- Managed style, lighting, tone, and artist weight inheritance
- Engineered cascading parameters across multiple layers
- Used version tracking for reproducibility and creative iteration
- Supported structured generation workflows for resumes, narratives, DBT matrices, and healthcare artifacts
3. Taxonomy + Tagging Infrastructure
- Built tagging and error classification systems for:
- LLM output evaluation
- Prompt logic chains and chain-of-thought reasoning
- Version drift and tone decay
- Maintained multi-layer tagging strategies in Obsidian (e.g., memory tags, Codex entry taxonomies)
4. Real-World Application Domains
- Healthcare: Structured letters to doctors, medication logs, diagnostic argumentation
- Mental Health: Applied DBT framing to narratives, logged emotional vectors, built symptom matrices
- Career Development: Applied model logic to resume building, tone drift correction, prompt polishing
- Creative Narrative: Evaluated longform stories for coherence, stylistic bleed, and rhetorical logic
🔧 Skills & Tooling