Resume

AI/ML engineer with full-lifecycle experience in architecture, deployment, and orchestration of containerized ML pipelines, GPU-accelerated systems, and agentic tooling delivered to production for real clients.

Ethan Muhlestein

AI/ML Engineer

ehmuhlestein@gmail.com 801-318-3153 Provo, Utah LinkedIn

Background

I build systems that run in production. Three years of applied engineering across data analytics systems, GPU-accelerated pipelines, numerical optimization, computer vision, and agentic AI tooling. My work sits at the intersection of data analytics, ML/LLM models and the infrastructure needed to deploy them, including containers, APIs, cloud setup, and automated workflows. Contractor work building document processing and OCR pipelines for enterprise clients. Internship work delivering choice modelling and optimization tools to data analytics clients. Personal lab running GPU workloads for 3D reconstruction, self-hosted services, and custom AI tooling. Python-first: FastAPI, Numba JIT, LLMs, Langchain, Docker, GCP, CUDA.

Experience

Contractor

TopDial Solutions

Mar 2025 - Present

Production ML and document processing pipelines for a large non-profit client.

Document Indexing Pipeline

  • Architected multi-container system with a Next.js frontend, Python FastAPI backend, and Firestore database
  • Built GPU-accelerated ML model for image illumination correction prior to OCR, significantly improving extraction quality through Google Document AI
  • Designed integrated human review workflow for quality control at scale
  • Stack: Python, FastAPI, Next.js, Firestore, Docker, GPU compute

Historical Record Metadata Pipeline

  • Built LLM inference pipeline running on cloud GPUs for structured metadata extraction (dates, record types, key fields) from scanned historical documents, targeting millions of images
  • Validated extraction accuracy at smaller scale, iterating on prompt design and preprocessing before full-volume deployment
  • Stack: Python, vLLM, Llama.cpp, OCR, cloud GPU provisioning with local LLM models: Chandra 2, Qwen 3.6, and Gemma 4.

Data Analyst / Engineer Intern

The Analytics Team

Jun 2023 - Present

Choice modelling and data analytics for enterprise clients.

  • Rebuilt legacy Excel VBA genetic optimizer into a containerized Python backend (FastAPI, Numba JIT, NumPy) deployed on Google Cloud Run with autoscaling. Runtime dropped from 3-5 minutes for 5-6 generations to 15-20 seconds for 12+ generations. Google Sheets frontend via Apps Script with API-triggered execution.
  • Developed agentic survey processing workflow using Python and LangChain for interactive survey data analysis
  • Delivered perceptual driver maps and choice model outputs to clients with product attribute correlation against purchase likelihood, share of preference, and revenue optimization
  • Extracting data from a custom structured Excel spreadsheet into an AI readable format
  • Stack: Python, NumPy, Numba, FastAPI, LangChain, GCP Cloud Run, Google Sheets API, Docker, R, SPSS

Projects

3D Gaussian Splatting Pipeline

  • End-to-end containerized pipeline running image upload, then COLMAP for structure-from-motion, then Gaussian splat training, all NVIDIA GPU-enabled via Podman
  • Motivated by 6 years of 3D art and photogrammetry work. Entered multiple global render challenges (Pwnisher). Deep understanding of mesh geometry, procedural texturing, rendering, and spatial reconstruction pipelines. You can view my artstation here: https://www.artstation.com/ethanmuhlestein
  • Runs on a local GPU server with remote access through Tailscale mesh VPN
  • Stack: Python, COLMAP, Gaussian Splatting, Podman, NVIDIA CUDA

Agentic AI Tooling

  • Daily use of Pi Coding Agent harness with custom extensions, Claude Code, VSCode, Kilo Code, Cline, OpenCode
  • Comfortable evaluating and correcting AI-generated output.
  • Always tailoring a custom AGENTS.md per project with rules and context.
  • Custom OpenAI-compatible tool call pipelines and extensions for AI-assisted development workflows
  • LangChain/LangGraph.

Self-Hosted Infrastructure

  • Docker, ProxMox VE, Tailscale mesh VPN, remote IDE, container networking across on-prem and cloud environments

Education

Computational Data Science

Utah Valley University, Orem, UT

Coursework completed