$114M SOV Reconstruction & QA
What It Does
A $114M core & shell SOV had been converted from PDF to Excel with catastrophic data corruption — 121+ merged cells, concatenated line items, truncated descriptions, and missing entries across 42 trades. Used AI to parse the source document and build a Python correction script (1,575 lines) that mapped 129 specific fixes: 73 combined items needing splits, 34 wrong descriptions, 16 missing items, and 4 page header artifacts embedded in data. Automated ~75% of corrections, then completed a 5-hour manual QA pass cross-referencing three sources simultaneously. Delivered same day — reduced an estimated 3-4 day effort to ~7 hours.
Key Features
- 1,575-line Python script mapping 129 specific data corrections
- Automated 75% of cleanup across 42 construction trades
- Cross-referenced three sources in a 5-hour manual QA pass
- Delivered same day — reduced 3-4 day effort to ~7 hours
- Fixed 121+ merged cells, concatenated items, and missing entries
Why I Built It
A corrupted $114M SOV is a blocker for the entire project. Manual correction would have taken days. I used AI and Python to deliver a verified result in hours.
What I Learned
Learned how to combine AI parsing with programmatic correction scripts for large-scale data cleanup. Deepened understanding of SOV structures and construction trade categorization.