Why I Built Mintline

The Drawer I Kept Ignoring

Every month had the same little ritual. I’d pull open the drawer in my home office, dig past cables and manuals I should have recycled, and find a stack of receipts waiting for “future me.” By the time I finally sat down with my bank exports, the ink was fading, the totals were fuzzy, and the mental energy was gone.

I’m a developer. I love automation. Yet every month-end I was caught playing forensic accountant: scroll a CSV, squint at a receipt, guess which charge it was supposed to match. I’d put on a podcast, spend the better part of a Sunday, and wonder why the machines weren’t doing this for me already.

Six hours disappeared that way last tax season. Not because I procrastinated out of anger—because the task just never felt worth the context switch.

The “What If”

I’m building Mintline entirely with large language models. That sounds trendy, but it started as a very practical question: could I give an AI the inputs I stare at—bank transactions, receipts, patterns—and have it hand me clean matches?

I opened a new file, wired the first prompt, and asked the model to match receipts to transactions. It did... badly. So I kept iterating. Teach it that $24.99 and $25.01 are the same lunch after card fees. Explain that “Joe’s Coffee Utrecht” and “JOES COFFEE NL 8237” are twins. Let it see the timeline of a charge posting days after the receipt.

Weekends went by. I’d feed the model more context, let it critique its own match, tighten the instructions, and run another batch. The drawer started emptying faster. Twenty receipts matched themselves in a few seconds. Then fifty. Then almost everything.

At some point I realized I hadn’t done the month-end dance in weeks. The LLM handled it.

What Mintline Became

Mintline isn’t an app I babysit. It’s the teammate I wanted: an agent that understands my finances, remembers historical behaviour, and keeps asking the right clarifying questions until the job is done.

Here’s what it does today:

• Receives every receipt I forward and extracts the useful bits automatically
• Pulls transactions from my accounts and builds a running timeline of likely matches
• Explains its reasoning before it confirms anything (“matched to Utrecht coffee, amount within tolerance, dates aligned”)
• Flags the handful of edge cases it’s unsure about so I can make the call in seconds
• Learns from the decisions I make, so the next month is smoother than the last

The drawer is still in my office. It’s empty.

Why I’m Sharing This

Mintline wasn’t born out of rage. It was born from a quiet dissatisfaction with losing creative time to rote reconciliation. The more people I talk to—freelancers, founders, finance teams—the more I hear the same sigh. Everyone has their version of the drawer.

Large language models finally let us teach software the nuance we use when we reconcile by hand: habitual merchants, subtle fee drift, how a refund pairs with an original charge. Building Mintline with LLMs means I can hand the task to a system that learns with me instead of locking me into preset rules.

That’s the point. Not to chase hype, but to erase an annoying chore for good.

What’s Next

Mintline is still growing. Upcoming on the roadmap:

• Deeper integrations with accounting platforms so the matching flows straight into your books
• Smarter hints before transactions post (predictive matching based on merchant habits)
• Richer conversations with the agent—ask “Which invoices are still missing receipts?” and get an answer instantly
• Collaborative workflows so finance teams can delegate reconciliation to the AI and only review exceptions

If you’re someone who dreads the drawer, the shoebox, the inbox label called “Receipts,” I built Mintline for you. For the quiet relief of the first month you don’t have to do it yourself.

Let the LLM do the matching. You keep your weekends.
Start using Mintline and give the drawer a permanent day off.