Real builds. Real results. No promises required.
Every claim on this site rests on work that actually exists. Here are the projects, each told the same honest way: the problem, what I did, and what came out the other side.
A gigabyte of messy government data, made useful
The problem
NHTSA — the federal auto-safety agency — publishes a huge archive of vehicle complaints: over a gigabyte of raw text files. The insight buried in it is real, but no human can read a gigabyte of text, and the most interesting question — which part types fail, on which vehicles, in which patterns — isn't explicitly recorded anywhere in the data.
What I did
I wrote Python text-extraction tooling that reads the entire archive and pulls part-type failure signals out of the free-text complaint narratives — surfacing patterns the dataset never states outright. Then I rendered the results into clean HTML views, so the output isn't a data dump but pages a person can actually read and act on.
The result
A gigabyte of noise became a usable reference: failure patterns by part type, visible at a glance. And a demonstration of the core premise — a precise question plus grown-to-fit tooling beats expensive platforms and guesswork.
Systems that run forever with no per-use AI cost
The problem
The lazy way to "use AI" is to call a model for everything — and pay a little every single time, for answers that can vary run to run. For truly repetitive tasks, that's the worst of both worlds: recurring cost and unreliability.
What I did
I built local Python systems — deterministic scripts, developed with AI's help but running without it — for the tasks where the rules never change. The AI's contribution was the build, not the runtime.
The result
Tools that cost nothing per use, produce the same correct answer every time, and keep working with no subscription attached. Knowing when to use this pattern instead of a model call is one of the most valuable judgment calls I bring to client work.
AI-assisted workflows for the tasks that genuinely need AI
The problem
Some recurring work isn't rule-based enough for a plain script — it involves messy input, language, or judgment at the edges. That's where AI actually earns its keep.
What I did
I set up AI-assisted workflows for exactly those tasks — with the AI handling the messy middle and the human keeping the judgment, the checks, and the final say.
The result
Recurring hours reclaimed — and a clear, honest boundary line between "script it," "AI-assist it," and "keep it human." Drawing that line correctly is most of the game.
New tools for the automotive aftermarket
The problem
The automotive aftermarket runs on hard-won practical knowledge — but some of the tools the industry could obviously use simply didn't exist.
What I did
Building on the NHTSA data work and everything learned along the way, I built brand-new tools for the aftermarket from scratch.
The result
Working tools that didn't exist before — created not by a software company, but by someone who started a year ago knowing almost nothing about code. That trajectory is the product.
The unglamorous stuff, automated.
Not every win is a gigabyte of government data. Most of them are ordinary workday tasks that quietly stopped costing hours. Names and employers withheld on purpose — the patterns are what transfer to your business.
Repetitive forms and orders, filled by script
The problem
The same web forms and order screens, filled in by hand, over and over — identical steps, identical rules, real hours lost every week to typing what a system already knew.
What I did
Had AI help me write scripts that fill and submit those repetitive forms and place the orders — same rules, every time, no copy-paste marathon.
The result
A weekly chore became a button press. Zero transcription errors, and the person who used to do the typing now checks the output instead of producing it.
The mega-spreadsheet that stopped crashing
The problem
A massive, formula-driven report workbook — so many live formulas that updating it would routinely crash the file. Maintaining it was a job in itself; trusting it was a leap of faith.
What I did
Replaced the formula forest with a script: read the source data, compute everything in code, and write out a clean, ready-to-share report.
The result
Updates that took a fragile afternoon now run in seconds and never crash. Same numbers, same layout — none of the drama.
Prospecting that finds its own opportunities
The problem
Finding gaps in a customer's catalog — the products they could carry but don't — meant hours of manual comparison, and then more hours building a presentation to show them.
What I did
Built an automated pipeline that compares catalogs against available data, flags the gaps worth talking about, and generates the PowerPoint deck for the conversation — automatically.
The result
Prospecting prep that took most of a day now takes minutes, and every meeting starts with a deck built from evidence instead of guesswork.
An inbox that sorts itself
The problem
Email management eating the edges of every day — the same routing, filing, and follow-up patterns, performed by hand, forever.
What I did
Built Power Automate flows — using tools the business already paid for — to route, file, and surface what actually needs a human reply.
The result
The inbox went from a task list somebody else writes to a filtered queue. The repetitive handling happens on its own; attention goes only where it's needed.
Every one of these follows the same pattern: a script for the repetitive spine, AI only at the messy joints. That pattern is teachable — it's exactly what the training covers.
Your repetitive work is somebody's future case study.
Preferably yours. Let's talk about what's eating your hours and whether a small, honest tool can take it off your plate.