Agents In Use
What’s new todayLast update
108 use cases added today

Codex to review a year of receipts before tax export

Matthew Vroom used Codex to clean up a year's business expense records before tax preparation. His starting folder contained unlabeled PDFs, receipt photos and emails, and it was incomplete. The job was both organizing the evidence and identifying what might be missing.

Source published linkedin.comAgent: Codex
FROM THE SOURCE

Original prompt

Shared by the source author. See source notes for context.

If confidence is still weak, mark the document for review instead of storing a junk vendor.

Evidence at a glance

Sourced account
Source author
Matthew Vroom
Original account
Read the source ↗
Published
Source link checked
Reported result
He demonstrated a local receipt review dashboard and Schedule C preparation exports after a reported twenty five to thirty minute agent run.
Human involvement
Vroom demonstrated source inspection and review tabs; final tax filing and correctness of every category were not confirmed.

A linked account supports attribution. It does not mean we independently reproduced its result.

Report an issue with this caseOpen correction email

Opens your email app with the case link included. Nothing is sent automatically.

How the agent did it

  1. Point Codex at the mixed receipts folder and process supporting statements.
  2. Inspect parsed receipts, duplicate candidates and failures in the local dashboard.
  3. Review Schedule C and supporting exports before using tax preparation software.

Matthew Vroom used Codex to clean up a year's business expense records before tax preparation. His starting folder contained unlabeled PDFs, receipt photos and emails, and it was incomplete. The job was both organizing the evidence and identifying what might be missing.

He pointed Codex at the folder and had it parse receipts, rename files, build a ledger and cross reference statement data. After about twenty five to thirty minutes, he demonstrated a local dashboard. Receipt views exposed the source document, vendor, amount and extracted details instead of hiding the evidence behind a total.

A review tab separated duplicate candidates, unsuccessful parsing and uncertain expense records. Reports included a Schedule C entry summary intended for TurboTax, with supporting expense details. Vroom described these as preparation outputs; the video does not confirm that a return was filed or that every proposed category was correct.

The linked published prompt is specifically a ReceiptSweep demo workflow. It keeps original source files unchanged, reuses a designated runtime and asks for conservative review flags instead of weak vendor guesses. That distinction matters when copying the prompt: it is a documented demo instruction, not a promise that an unfamiliar receipts folder can be accepted without inspecting its review queue.

Prompt
Original prompt

Source: www.linkedin.com · Matthew Vroom

Frequently asked questions

How could extraction be checked?

The receipt view showed the original document alongside vendor, amount and other extracted details.

What did the review tab separate?

Duplicate candidates, unsuccessful parsing and uncertain expense records were surfaced for inspection.

Is the published prompt the exact live tax run?

The linked instruction explicitly describes a ReceiptSweep demo workflow with its own source and runtime paths.

More details

About this account

Codex parsed and renamed expense evidence, constructed a ledger, reconciled statements and generated a local review and export interface.

Firsthand video transcript and linked prompt opened. Prompt excerpt is verbatim from the published ReceiptSweep demo instruction, not claimed as the exact instruction used on all original business records.

Credit: Matthew Vroom

Source published

Added to Agents In Use

Source checked

Tools and topics

Codex · Local dashboard · PDF · CSV · TurboTax

Receipt review · Statement reconciliation · Schedule C · Local dashboard

Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.