Fuel Card Reconciler
Import fuel-card transactions, compute MPG per vehicle, and auto-flag fraud and waste - over-tank fills, off-hours charges, and duplicates - so a fleet manager clears or investigates exceptions before anything posts.
A private internal tool that ingests fuel-card CSVs, computes MPG per vehicle, flags anomalies, and routes flagged charges through a manager approval gate before they post - with a full audit trail and a clean cleared/exception CSV export.
Before you start
- A free Supabase account
- A free Vercel account
- A free Resend account
- A recent export of fuel-card transactions as CSV
The problem this kills
Fuel cards leak money quietly. A driver fills a personal truck, a station double-charges, someone tanks up at 2 a.m. on a Sunday, or 40 gallons go into a 25-gallon tank. By the time the statement hits accounting, it's a wall of rows nobody has time to question - so it all just posts. Multiply a few bad charges across a fleet and a billing cycle, and you're funding waste and outright fraud every single month.
The honest problem is that catching this by eye is impossible at scale. You'd need to know each vehicle's tank capacity, each vehicle's normal MPG, who was assigned to drive it, and what "off-hours" means for your operation - and then check every line against all of that. No one does it. The charges win by sheer volume.
What you'll build
A private web app where you drop in a fuel-card transaction CSV and the tool does the checking for you. It matches each charge to a vehicle and driver, pulls the latest odometer reading, computes miles-per-gallon, and runs a battery of fraud and waste checks. Clean transactions sail through. Suspicious ones get flagged with a plain-English reason and land in a review queue, where you - the fleet manager - decide write-off vs. investigate. Only after you approve does anything clear. You walk away with a cleared file ready to post and an exception report you can hand to anyone.
What's inside the Implementation Plan
The plan is a complete, paste-and-go runbook for an AI coding agent (Claude Code). You don't write code - you paste, answer questions, and approve.
It opens by interviewing you about your fleet - your fuel-card provider and the exact columns in their export, how vehicles and drivers are identified, your tank capacities, your normal MPG ranges, and what counts as "off-hours" for your operation. It uses your answers to tailor the data model, the anomaly thresholds, and every later build step, so you get a tool shaped to your fleet, not a generic demo.
From there it walks you through, step by step: standing up the database, building the secure login, importing your CSV with duplicate protection, computing MPG, running the anomaly checks, building the manager review-and-approve queue, wiring up email alerts, and exporting the cleared and exception files - each step ending in a ready-to-copy prompt.
The governance it includes (this is the point)
This is not a throwaway script - it's an internal tool you can trust with money decisions:
- Login so only your team can open it.
- Row-level security so each organization only ever sees its own fleet and charges.
- A complete audit trail - who imported what, who approved or rejected each exception, and exactly when.
- A hard human-in-the-loop gate - the tool flags and drafts, but nothing clears for posting until a manager reviews and approves.
- Duplicate guards - the same charge (card + timestamp + amount) can never be imported and processed twice.
Who it's for
Fleet managers who own the fuel budget and the finance teams who post the charges. If you currently reconcile fuel cards by squinting at a statement - or not at all - this turns it into a quick, defensible review.
You've got this. Paste the first prompt and let the agent interview you.