M.S. Applied AI, Neumont University
Ten years of software engineering, now focused on RAG, LLM orchestration, and the data pipelines that make them trustworthy.
I spent a decade writing production software. Now I build the AI that runs on it: retrieval pipelines, fine-tuned models, and the evaluation that proves they work.
I'm an AI-focused engineer based in Ogden, Utah. My background is classic software engineering: backend services, SQL systems, CI/CD pipelines. For the past several years I've rebuilt that foundation around AI, with RAG pipelines, multi-model orchestration across Claude, Gemini, and ChatGPT, local LLM hosting, and the ML data work that makes any of it trustworthy.
My M.S. in Applied AI pushed me deep into the parts a bootcamp skips: ETL design, data sanitization, model training and evaluation in TensorFlow and PyTorch, and the statistics behind knowing whether a model actually works. I pair that with real production responsibility, validating a live AI agent at Blytz.
I build with Claude Code as a first-class part of my workflow. Not to skip the engineering, but to move faster through it while staying accountable for every architectural decision underneath.
Neumont University. Model training and evaluation, RAG architecture, ETL and data sanitization.
Weber State University. Graduate-level data science coursework as an undergraduate.
Graduate capstone
An open-source, self-hosted AI campaign companion for tabletop RPGs. A LoRA fine-tune of Mistral 7B, hybrid retrieval over the user's own campaign records, and an evaluation harness that measures both. No external AI APIs.
Live and open source, AGPL-3.0
A session-journal corpus across 12 tabletop game systems (fantasy, sci-fi, horror), with cleaning and formatting pipelines that made fine-tuning trustworthy.
LoRA fine-tune of Mistral 7B Instruct, merged and quantized to Q4 GGUF. A 4 GB model that runs on consumer hardware, published on Hugging Face.
Semantic search (pgvector) fused with full-text keyword search via Reciprocal Rank Fusion. Embeddings handle paraphrase; keywords rescue the rare proper nouns they blur.
The retrieval threshold is set by a labeled evaluation sweep, not guessed. When nothing relevant exists, the app labels its own reply as ungrounded instead of inventing.
In-app comparison of the fine-tuned and base models with semantic-similarity scoring, plus a standalone retrieval benchmark. Quality is measured at both layers.
Dockerized FastAPI, React, Postgres, and Ollama with token streaming and a per-answer retrieval trace. Every reply shows which records it drew from and why.
RAG quality is a data quality problem.
Multi-agent AI team platform
A self-hosted crew of AI personas (product manager, engineers, QA, DevOps) working one shared Kanban board inside a sandboxed git workspace. They pick up tasks, write the code, and open branches for human review, with any model provider, spend caps, and a full audit trail.
Coming soon to iOS
Private voice notes with on-device AI
Records, transcribes with Whisper, summarizes, and pulls out tasks on iPhone and iPad. Every step runs on the device's own chip, even in airplane mode.
Sanitization, ETL, training, and evaluation in Jupyter and ML.NET, deployed as a working image classifier.
GitHub Actions pipelines that compile and deploy production builds straight to a live site.
Blytz
Common Thread
Salt Lake Community College
Hill Air Force Base, Instructure, Microsoft, Control4
Years as an AI Trainer at Salt Lake Community College taught me how to make RAG, LLM orchestration, and applied ML approachable. I bring that to workshops, mentorship, and training for teams new to them.
I build AI capabilities into existing software: RAG search, LLM orchestration, evaluation pipelines, and the data infrastructure that keeps AI features trustworthy in production, not just in a demo.
Open to AI development and teaching
Based in Ogden, Utah. Remote or local.
