AI/ML Engineer building LLM fine-tuning pipelines, RAG systems, and evaluation tooling.
Multi-stage fine-tuning system for creative text generation: LoRA/QLoRA fine-tuning, custom reward models, GRPO optimization. The public components:
- prosody-judge — LLM-as-a-judge evaluation layer. Scores both reasoning traces and final lyrics across 8 rubric dimensions, with 3-run self-consistency, consensus flagging, and structured JSON output used as GRPO reward signal.
- songcraft-rag — knowledge-retrieval layer. 45-document, 2,376-page songwriting & music theory corpus indexed into 7,370 chunks; local ONNX embeddings, ChromaDB, cited answers via CLI and FastAPI.
- repo2GPT — turns any repository into an LLM-ready repo map plus consolidated code bundle. Language-aware summaries for Python, JS/TS, Go, Rust, Ruby, and PHP.
- pixel-perfecter — reconstructs clean, grid-aligned sprites from diffusion-generated pixel art. Grid estimation and color-cell reconstruction core, PySide6 GUI, batch CLI.
Python · PyTorch · Hugging Face · LangChain · ChromaDB · OpenAI/Anthropic APIs · FastAPI · React · Node.js · MLX · llama.cpp
Azure AI Engineer Associate certified · Based in WA · Open to AI/ML engineering roles