Braeden K.

Every number traces to a source, and every pipeline tells you when it's wrong. I specialize in messy data reconciliation, complex document extraction, and production Python pipelines where dropping a single row is not an option. Whether you have legacy mainframe PDFs, conflicting multi-platform exports, or thousands of unstructured records, I turn fragmented sources into clean, verifiable deliverables. What I build: • Complex Data Extraction: 9,915-page legacy mainframe PDFs, scanned tables via OCR, and rate-limited web sources parsed into structured schemas with 100% row accounting. • Records Reconciliation & Deduplication: Combining overlapping exports (WhatsApp, email archives, CRM/ERP dumps) into clean, chronological, audit-ready databases. • Financial Models & Budget Workbooks: Cost and billing-rate models built in Excel — overhead pools reconciled across GL accounts, labor allocated per person, every output traceable to the input that drives it. • Defensive Pipelines & Automation: Unattended daily scrapers and ETL jobs built with automated validation rails, checksums, and loud failure alerts—never silent errors. • Database Migrations: Microsoft Access front ends over SQL Server, including guarded column-type migrations that refuse to run against the wrong database and roll back if verification fails. • Clean Reports & Dashboards: Turning raw CSV/Excel dumps into interactive HTML dashboards or automated client-facing charts. Recent Track Record: • Multi-Source Forensic Reconstruction ($900 Upwork project): Reconstructed 7,600+ fragmented communications across 7 platforms into a single searchable, blind-review interface (5.0 ★). • Invoice Backup Automation (ongoing client): Five invoices, 383 assembled pages, 155 labor rows and 162 expense lines matched to source documents — signed-timesheet ingestion, scanned-receipt OCR, purchase-and-return credits, and mid-cycle invoice revisions. Every placed page traces to a source file and hash. • Commercial Billing-Rate Model (ongoing client): 244 formulas, zero errors. Overhead pool reconciled across 40+ GL accounts, billable and non-billable hours allocated per employee, cash recovery separated from operating profit. • Public Records Intelligence Engine (GAScout): Parsed 409,142 county tax records from a 9,915-page fixed-width PDF with zero dropped rows. • Live Market Observatory: 9,055 storage facilities across 7 national brands, collected daily for 57 consecutive days — 128,354 advertised units tracked, unattended, with run-level reconciliation. Collection methodology and findings published publicly. • Published, graded forecasts: Three predictions committed to a public git repo before a quarter closed, with falsification rules written in advance, then graded in public against the company's own 10-Q. One hit, one missed. The miss was published with its arithmetic. Fixed price or hourly. Send me a sample of the raw input and what you need out of it—I'll tell you honestly what's straightforward, what's tricky, and give you an exact price and timeline.