Last summer my long-held position at a large tech company ended due to layoffs and budget cuts related to AI expenditures. That should sound familiar to a LOT of you, unfortunately. For me it was both a blessing and a curse.
The Blessing
I had been dealing with some long-running chronic health issues and was making NO progress. I was actually thinking about taking disability leave to focus more on my health when the layoff notification came out of the blue. I was shocked, a bit, but not rocked: it was unexpected but also a bit of a relief. Work stress was suddenly GONE. Family and everyday life stress were still a thing but that stress was much more manageable.
The layoff came with a nice severance package, which is pretty standard for the tech world these days. Since I had been at my job for a LONG time, wow, did it help. Then came unemployment compensation, which added another layer on top. My wife and I have been very conservative with our money over the years, which started to pay big dividends-
1. Our house is paid off: no mortgage/rent payment, just property taxes. Which are actually higher now per month than our original mortgage but that's another story.
2. NO DEBT: this is HUGE. We have zero debt. None. Nada. Cars are paid for. House is paid off. No college debt.
That leaves us with very low overhead that easily lives within our savings envelope. Our biggest expense each month is healthcare (shocking, right?) but it is still sustainable until I find a new job.
The Curse
Most of my daily life was defined by my job. Now that was gone and I had nothing to replace it, which worked great for the first 6 months as I recovered my health, but became a worry around last January. Lunches out with former coworkers only get you so far. I can't state how big of a worry this became. What would I do? What will be my daily purpose?
The AI Turnaround
One day, last fall, a friend asked me to lunch to talk about my job search. He meant well and his advice was sound: don't wait. Keep looking, put yourself out there, and do something to keep your skills fresh and relevant. I tried to follow his advice but it just didn't really gel for a few months.
Then, a few months later, that same friend asked me if I wanted to join a group of people from our church to learn about AI, play around with it, and be kind of an unemployed support group. I absolutely jumped at the chance. We, the 6 of us, started meeting every Friday morning almost immediately.
Each week we spent an hour talking about how to use AI in our lives and how we might have used it in our previous roles. We were challenged to find a "passion project" or a way to use AI in a hobby or other way that we might put on a resume. The second hour was spent with a guest speaker, typically someone that is a friend/coworker of someone in the group, that works with AI either as their primary role or as a big part of their job: think AI engineer versus an Engineer, Program Manager, or People Manager who use AI extensively in their daily work.
It really woke me up. I started using Claude and ChatGPT to complete a long list of AI competency courses. I used it to brainstorm several project ideas and was coming up a little short. After quite a few fits and starts, my passion project came into focus and I named it Fuelytics.
Fuelytics
It all started with my father-in-law. Back in the 70s, during the OPEC oil embargo, he, along with a lot of other people, started logging all of his fuel purchases to keep track of costs and fuel efficiency. It was clunky (just paper and pencil) but it worked. He got my wife into it and then she got me into it after we married. The idea is simple: every time you purchase gas, log the date, odometer reading, cost per gallon, gallons purchase, and total price. And maybe add in an oil change or other important maintenance task along the way.
THE PROBLEM
The hard part was getting all of that data into Excel. This turned into HOURS of tedious work manually typing all that data into a spreadsheet. If I compare the ratio of time spent on the stats analysis versus the time doing manual data entry, it's like 10 to 1. If only there was an app that could automate this process!?!
I ruminated on this problem for over a decade. Yes, I could write an app, but I didn't have the time to write the code or, more importantly, the time to dedicate to learning how to write the code to be able to make a workable app. So the idea sat.
Fast-forward to today, where I'm thinking over what kind of project to do. Eventually I came around to all of this vehicle fuel data from back in the day. What might I do with it? Could I do something fun with the static dataset and answer some fun questions? Yes, but that's essentially what I did back in college and didn't really spark my interest.
What did spark my interest was an old question: what about that app idea? Could I create an app to automate the hard part of this process? So I did some research: I combed the Apple App store and Google Play store for fuel trackers. There are quite a few but all of them rely on manual data entry. Still not as easy as I would like. I would rather just point my phone at the pump, tap a button, point it at my odometer, tap a button, and be done.
And that's exactly what I built, but let's not get ahead of ourselves.
The Solution
So I paid for a Claude subscription and got to work with Claude Code. I picked up a "pro" subscription to start, since this is all on a very tight budget. I'm unemployed, remember?
I started with an extensive planning process. Headlines in the news about vibe coded apps causing havoc and exposing sensitive data to everyone really made me think about setting up clear guardrails and protections first, and then write the code. Working with Claude Code, I defined the data model, security policies, privacy policies, and operational model. Where would the data live? How do we secure it from the beginning and keep it safe, even though this is just vehicle fuel purchases we are talking about? I documented everything and used that as the basis for the development effort.
Within a couple of weeks I had setup a wire-frame website. The idea here was simple: create a working authentication model, secure website, initial data structure, simple user and data flow, and tie it all together. This turned out to be the hard part but it worked. I even signed up for a ChatGPT Codex account and had it do a security review of what I did with Claude Code. Then I added graphic design using Claude Designer and enough features to implement the basic version of the website: www.fuelytics.app.
That first version was simple and basic: create an account, create vehicle definitions, add fill-ups manually, and see a basic report of your activity. I had a few friends try it out just to see if it worked as well as I thought it did. Feedback was quick and straightforward: you just copied the manual fill-up apps. Where's the value?
And that was the next step: add the automation. Over the coming week I iterated on this idea with Claude by taking picture after picture, visiting gas station after gas station, tweaking prompts, and trying not to get frustrated. After a lot of trial and error we nailed it and the website went live: a fully working and complete web app that could invoke your phone's camera using the browser API, capture photos of the gas pump and vehicle instrument cluster, send it off to Claude for analysis, and return the resulting data to the user for verification. After the user verifies the data it is logged under the user account.
It even worked on mobile browsers-
And then came more feedback: what if a family wants to use this? Can you allow multiple people to see and add data to multiple cars at the same time (i.e. emulate a family with more than 1 driver and more than 1 car)? Turns out this was as simple as tweaking the data structure and adding a few more features. That sounds much harder than it actually was.
But could we turn it into a mobile app?
The Android App
From the very beginning, I was clear in my intentions to my AI PM that I wanted to make this a mobile app for both Android and iPhone. That led to some early design decisions that worked well in my favor: React.js is an amazingly flexible platform and ports very well when moving to a mobile native app. All of the functions and data flows were created with phones in mind.
The first Android builds came quickly. The testing and bug fixing took a lot longer. Once again, I was driving around to different gas stations doing "fill-ups" on my test accounts: drive to a gas station, use the app to capture the data from the last pump user (which is conveniently still on the pump when you leave and a new person, aka me, drives up), then capture the odometer of the car (even though I've driven a couple of miles).
And just to make the gas station owners happy I frequently went inside and bought a soda or snacks. I'm sure they were puzzled with my behavior of pointing my phone at the pump and then driving away without making a purchase.
The process worked well enough that I was able to eventually close enough bugs, do another code review, and close more bugs, to finalize the Android build and submit it to the Google Play Store. It will be published in the next couple of weeks.
Lessons Learned
This entire process was started on April 24, 2026, so we are 2 months and 2 weeks as of the day of this writing. Rapid prototyping, code writing, code review, and bug fixes can be done very rapidly with modern AI-assisted coding tools. So this whole thing has really satisfied my ADHD urges.
The hardest part was the idea. Once I had the idea the rest just kind of flowed: I have extensive experience with IT operations and data protection, which I put work on this project. It really helped to set the groundwork that is over-secure for what it actually does but runs efficiently enough for people to trust it.
Local AI on phones just isn't there yet. My first thought was to have the photo analyzed on-device. Google ML Kit and Apple Vision failed spectacularly: they were designed for basic OCR tasks like capturing text from documents or well-formed tables. Gas pumps are very messy and vary a lot from one company to another. How bad? Like 3-5% accuracy bad. Maybe someday they will get there but until then it's Cloud AI.
When it comes to AI, BE SPECIFIC. Just like any other code you write, it will do exactly what you tell it to do. If you don't give it enough detail or write something wrong, things go off the rails. With old school coding you just get weird results or your app/machine crashes. With AI you get hallucinations or it simply makes assumptions and doesn't tell you. My first wireframe prototype assumed I would be the "admin" who defined vehicles, "owner" users would be adding fillups to those vehicles, and "contributor" users would be read-only. When I discovered this mistake I went back and looked at our design discussion and realized that I had not been clear enough in my expectations. What resulted is the group-driven model we have now: multiple users can share a group of cars and add data. User roles are defined with "Owners" (define vehicles, invite other users to share their vehicles, and add fill-ups), "Contributors" (add fill-ups), and "Viewers" (read-only users).
What's Next
Since that first website release I have added Google and Apple sign-in, bulk data import of historical data, ability to add "events" such as an oil change or other maintenance event, and a few other minor tweaks like an app support form and navigational aids.
The next steps may include another go at on-device analysis (if Apple and Google release better local phone models), the ability to toggle between metric and imperial metrics (it's all US imperial for now), and even event reminders based on your maintenance events to remind about upcoming oil changes or regular scheduled maintenance.
The biggest thing I may add is a way to import all those hand-written pages directly into the app. That's years and years of historical vehicle data that I could finally start using. Maybe, if the handwriting recognition works the way I hope it does.
Wait... Where's the iPhone App!?!
Yes, iPhone is the elephant in the room. For all of my career I have been a Windows user. I have always admired Mac for their UI, stability, and ease of use but never really pulled the trigger on day-to-day use. And I've never been an iPhone user. Yes, those are just excuses. But the real reason is that I simply do not have ready access to an iPhone for testing purposes. Testing, debugging, and code iterations require frequent physical access to a device, especially with an app like mine that requires a camera. If it didn't require the camera I could probably get away with testing on virtual or cloud phones but that simply isn't an option. And buying an iPhone just for this project would set me back $350-500 which I can't afford right now (unemployed, remember?). So if any of you want to donate an iPhone (13 or newer) feel free to speak up.
And that's it. I am now a real app owner/operator running solo, for better or for worse. Hopefully this will reach a calm stead-state before a find a new job or this could get hairy quickly.
}B^)
No comments:
Post a Comment