Auction Screenshot Evidence Intake
Version 1.2.1 has 47 passing tests. Extracted drafts still need a person to correct and approve them.
An offline screenshot-to-CSV/JSON workflow that detects overlapping captures, preserves image provenance and requires review of extracted auction history and roster names.
- Why
- I wanted to reduce manual entry while ensuring each detected bidder or price could be traced to the actual screenshot.
How I built it
- I added local screenshot selection and Tesseract-based candidate text extraction.
- I kept history events separate from bidder-roster entries.
- I tested real auction screenshots and found that the first parser yielded no history rows.
- I revised the parser for actual layouts and added explicit reviewer decisions, overlap handling and guarded exports.
- To copy it you need
- Collect only screenshots you are allowed to analyze and test one fictional example before handling real evidence.
Try it yourself
Copy this prompt and paste it into ChatGPT or Claude to try it yourself.
I want to copy the screenshot reader from Jovan Diaz's What Are We Building? profile. It runs offline: it pulls candidate text from screenshots with Tesseract, keeps the bid history apart from the list of names, and makes every extracted item traceable to its screenshot. A person approves each item before anything is exported. His gotcha: it passed tests on invented screenshots, then found zero history rows on real ones, so he had to rebuild the parser. Help me start with one fictional screenshot I'm allowed to use. Begin by asking what my screenshots show and what I want out of them. Ask me for anything you need.
Tools I used for this project
ChatGPT: I used ChatGPT to draft, test, critique or work through the workflow.
Codex: I used Codex to help build or revise repository files and tests.
Python: I used Python scripts or validation tools to process and check structured outputs.
GitHub: GitHub holds all of my repositories: code, instructions, changes and review history.Tesseract OCR
