About
From audit files to working software.
I’m Daniel. I studied accounting, spent three and a half years in audit at Deloitte, then became a financial analyst at Conservice, a utility-billing company. Most of that work lived in spreadsheets, and I kept automating it.
That background is still how I work. An auditor doesn’t trust a number until it traces back to something real, and an analyst goes looking for the manual work hiding in a process. In April 2026 I started building the software I had wanted as an analyst, beginning with Chirp Utilities. Some of what I’ve built since runs a business; some is for my family, my neighbourhood and my congregation.
How AI fits in.
I build with AI coding agents, mostly Claude Code and OpenAI’s Codex. They write a great deal of the code, and I won’t pretend otherwise.
My part is the rest: deciding what gets built and for whom, writing the rules the agents work under, reading what they change, and refusing to call anything finished until a test or an audit shows it works. The projects here are described by what they do and where they stand, not by numbers I can’t back up.
What I focus on.
Product thinking
Starting from who the software is for and the job it has to do, then cutting what doesn't serve it.
Workflows
Finding the manual steps hiding in a process, the kind I spent years automating in spreadsheets, and building them in properly.
Responsible AI
Using models where they help, labelling what they guess, and keeping a person in charge of the decisions that matter.
Testing
Tests and audits that prove the important parts work, before anyone calls a feature finished.
Security and privacy
Data kept apart per account, secrets kept out of code, and personal details kept away from models wherever possible.
Practical apps
Small, useful tools for a household, a family or a congregation, built to be used every week.
The rules I build by.
- 01
Money reconciles to the cent.
In Chirp a bill can't be finalized until resident charges plus the owner's share equal the provider bill. Farm Store keeps every amount in whole cents and won't refund twice.
- 02
A claim is not evidence.
An agent saying the work is done doesn't count; a passing test does. Ministering Scheduler went through three adversarial audits, and each weakness was reproduced by a failing test before it was fixed.
- 03
A person makes the call.
Where software guesses, a person confirms. Chirp flags uncertain bill fields for review, and the Communications Center never lets an AI guess look like a confirmed fact.
- 04
Privacy is a design input.
The Communications Center replaces names with tokens before anything reaches a model. Local Reader runs its speech model on the user's own GPU, so the text never leaves the machine.
Before this
Audit at Deloitte, financial analysis at Conservice, and years of automating the spreadsheets in between.
Full work history
- 2026 —
Founder, Chirp Utilities
Building with AI
Utility billing software for property managers, and everything else on this site.
- 2023
Financial Analyst
Conservice · utility billing
Annual expense forecasts with business leaders, and process automation. Found 15+ billing issues that started the recovery of $23k a month and $100k in back payments.
- 2020 — 2023
Senior Audit Assistant
Deloitte · Delivery Center
Automated 15+ recurring Excel reports with Power Query and PowerPivot, saving more than 100 hours a year. Taught Excel training sessions, wrote migration macros, and worked on a global RPA project.
- 2018 — 2019
Accounting, Special Projects
Center of Excellence for Higher Education
Streamlined spreadsheet processes, saving about $10k a year, and reconciled monthly inventory reports.
- 2019
BA, Accounting
Brigham Young University–Idaho