AI and data systems, built to ship.

I'm Harrison Smith, an AI and data engineer. I scope, build, and ship production systems, not demos.

Email me See my work
Deckhand live embed · auto-refreshing
  • Retrieval-augmented generation
  • Multi-agent orchestration
  • BigQuery and ETL
  • PostgreSQL and pgvector
  • Ollama and local LLMs
  • Evaluation harnesses
  • FastAPI
  • GitHub Actions CI/CD
  • LoRA fine-tuning

A reporting platform that replaced Tableau.

Explore all 15 apps

15 internal BI apps I've built and shipped, reproduced here at a high level: the real data, and most of what each app actually does, stay internal. What's below is invented companies and numbers standing in for the real thing, enough to show how Deckhand, the deck builder, embeds live views from the rest of the platform and keeps them refreshed.

15apps shipped and in active use
44legacy Tableau workbooks retired
1deck builder that auto-refreshes from the rest
3AI coding agents in daily use: Claude Code, Grok, Codex
Flagship build

Deckhand

A deck builder that embeds and auto-refreshes live views from every app below.
Leadership · Ops

Pulse

The company-wide dashboard leadership checks every morning.
Data platform

Atlas

A visual map of the warehouse, cross-checked against the AI agent's own memory of it.
Sales

Pipeline

Rep and manager sales dashboard: pipeline, win/loss, activity, commissions.
See the impact

I spent a decade writing production software. Now I build the AI and data systems that keep businesses running on it.

I'm Harrison Smith, an AI and data engineer based in Ogden, Utah. My background is classic software engineering: backend services, SQL systems, CI/CD pipelines. The last 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 data engineering that makes any of it trustworthy.

Day to day, I'm a Business Intelligence Analyst at a venture-backed payments company, where I build the BigQuery models and dashboards Finance and Risk run on, and own the evaluation framework for their production AI agent. I bring that same production discipline to independent contract engineering through HSmithDev LLC.

That role is also where the reporting platform below comes from: 15 internal BI apps, replacing a legacy Tableau stack, built with Claude Code, Grok, and Codex on top of a data warehouse I know end to end. I've also contributed training data and review to the company's internal AI agent, which gives those apps natural-language context about that warehouse.

I also help non-technical business users build their own apps on that same platform, through a framework I helped define and build. I train people on prompting, Claude Code, and other AI agents, connected over MCP to our internal AI agent, so building and deploying a new app isn't something only engineers can do.

My M.S. in Applied AI pushed me deep into the parts a bootcamp skips: model training and evaluation, ETL design, and the statistics behind knowing whether a model actually works. I build with Claude Code as a first-class part of my workflow, to move faster through engineering, not skip it.

10+ years of combined technical experience
More projects and source on GitHub
2026

M.S., Applied Artificial Intelligence

Neumont University. Model training and evaluation, RAG architecture, ETL and data sanitization.

2022

B.S., Computer Science

Weber State University. Graduate-level data science coursework as an undergraduate.

What I build.

Three areas I go deep in, whether I'm joining a team full-time or brought in as a contract engineer.

How I work

AI feature development

RAG & orchestration

I design and ship retrieval pipelines, multi-model orchestration, and agentic tool-use systems, then build the evaluation harness that proves they work before your users find out otherwise.

  • RAG architecture and hybrid retrieval
  • Multi-model orchestration (Claude, Gemini, OpenAI)
  • Agent evaluation and human-in-the-loop review
  • Local LLM hosting (Ollama, KoboldCpp)

Currently own the evaluation framework behind a production AI agent at a venture-backed payments company.

Data engineering and BI

ETL & BigQuery

I build the ETL pipelines and BigQuery data models underneath the dashboards your finance, risk, and product teams actually trust, then document every join so the next engineer trusts it too.

  • BigQuery data modeling
  • ETL pipeline design
  • SQL (T-SQL, MySQL, PostgreSQL)
  • Data validation and reconciliation

Validated roughly 150 table relationships in a production BigQuery data lake, now restructuring it into a medallion architecture.

Full-stack and backend engineering

APIs & CI/CD

Backend services, REST APIs, and GitHub Actions pipelines, scoped and shipped end to end. I own reliability on teams that don't have a dedicated platform engineer.

  • REST API design
  • GitHub Actions CI/CD
  • Google Cloud and Microsoft Azure
  • PostgreSQL and MongoDB

Ongoing backend and SQL systems work for Common Thread's clients since 2024.

RAG quality is a data quality problem.

Orchestration can't compensate for a bad chunking strategy, and a well-trained model is worthless if the interface buries the answer. I learned that building Lorekeeper, my open-source RAG project, end to end, and it shapes how I scope every AI engagement since.

How I work.

  1. Scope

    I start from the actual requirements, not a generic discovery deck: scope the system, flag the risks, and lock a plan before any code gets written.

  2. Build

    I build in the open: working increments, real commits, and Claude Code as a first-class part of the workflow, not a replacement for engineering judgment.

  3. Ship

    Production means production: CI/CD, monitoring, and documentation, not a demo that only works on my laptop.

  4. Support

    I stay reachable after launch. On contract work that means a support window; on a team it means I don't disappear once something ships.

Personal and open-source builds.

Alongside the reporting platform above, proof for the services above.

Ogden Agents

Open-source, multi-agent coding UI

A free, fully open-source (MIT) local app, in the spirit of herdr, for running Claude Code, Antigravity, Codex, Grok, or a local model from one browser UI instead of a terminal per agent. Permission cards before every command or edit, BMad planning and a live ticket board, and a UI you can theme down to the color.

  • Multi-agent
  • Permission cards
  • Open source (MIT)
  • UI theming
Ogden Agents' chat view: a project sidebar showing which chats are working or waiting, beside a permission card where Claude Code asks to run npm test.

In active build

PrivaSuite

Local-first AI apps for Mac and iOS

A family of on-device apps that never phone home: a desktop assistant for reading, workflow capture, and meetings, a PDF reader with synchronized narration, a meeting recorder that transcribes and summarizes on-device, and a study companion that turns notes into flashcards. One shared on-device engine, nothing sent to a server.

  • PrivaAssist
  • PrivaPage
  • PrivaMeet
  • PrivaStudy

Lorekeeper

Graduate capstone, open source

A self-hosted AI campaign companion for tabletop RPGs: retrieval over your own journal, grounded answers, and a fine-tuned model, so nothing it tells you is hallucinated lore.

ML classification pipeline

Sanitization, ETL, training, and evaluation in Jupyter and ML.NET, deployed as a working image classifier.

CI/CD for ThunderBeast.Studio

GitHub Actions pipelines that compile and deploy production builds straight to a live site.

ContractorCRM

Personal project, source private

A multi-tenant CRM, AI reporting engine, and assistant sharing one backend, with agentic AI execution centralized behind a dedicated MCP server.

AI book & video generator

Personal project, source private

Give it a premise and it writes a script or a book, keeps character art consistent across shots or pages, voices every line, and cuts the result into a video or a print-ready book. Entirely local models, no cloud APIs.

Learn Nepali

iOS & Android, in development

A Nepali-language learning app with spaced-repetition practice, a content-authoring studio, and AI-assisted lesson generation.

More on GitHub

Where this has been applied.

Full history on LinkedIn

Business Intelligence Analyst, Data Intelligence & AI

A venture-backed payments company

Nov 2025 - Present

Build the BigQuery models and dashboards Finance, Risk, and Product run on, ship the reporting platform that replaced its Tableau stack, and own the evaluation framework for its production AI agent.

AI & Software Engineer, Technical Consultant

HSmithDev LLC

Feb 2025 - Present

Scope, build, and ship software for businesses of all sizes, owning the full lifecycle from requirements to production. Consulting and AI workflow training for organizations including the PROG Development Center and Acord.AI, and subject-matter expert for Western Governors University's B.S. in Data Analytics with AI.

Software and IT Services

Common Thread

Mar 2024 - Present

Backend services and SQL systems for a variety of clients, owning reliability with no platform team behind me.

Data Analyst Coordinator & AI Workforce Trainer

Salt Lake Community College

May 2023 - Nov 2025

SQL and JavaScript dashboards for stakeholder reporting, plus lead instructor for "AI @ Work," a workforce training series on prompt engineering and practical AI adoption.

Earlier technical roles

Instructure, Control4, Education at Work

2015 - 2021

SaaS support for the Canvas LMS, IT systems administration, and Microsoft Office support.

Open to full-time roles and contract work

Let's talk about what's next.

hello@harrisonsmith.ai
LinkedIn GitHub

HSmithDev LLC. Based in Ogden, Utah. Remote or local.