Originations & Watchlist Platform

Client: Uplift Capital • Industry: Real Estate Credit / Data Engineering • Completed: June 1, 2026
Python Data Engineering Public Records ETL PDF Parsing

Challenge

The public records that drive real estate credit origination — county clerk and court filings and related public notices — are scattered across dozens of incompatible county systems. Each county publishes through a different portal, in a different format: scanned PDFs, HTML listing pages, fixed-width data exports, and vendor platforms that change without notice. There is no common schema, no shared identifier, and no reliable way to tell what changed since yesterday. Assembling a clean, current origination watchlist by hand across multiple counties is slow, error-prone, and impossible to audit — exactly the kind of reusable analytical infrastructure Uplift Capital’s approach is built to centralize.

Solution

Designed and built a deterministic, modular sourcing engine that fetches, parses, normalizes, and de-duplicates county public records across four Texas counties into a single daily watchlist — with every record traceable back to its saved source document, and a human making every origination decision.

  • Source-adapter architecture — one self-contained module per county and record type, so a broken or relocated source degrades only its own section and adding a new county is configuration plus a single module
  • Multi-format parsers for scanned PDFs, HTML listing pages, and fixed-width county data exports, with explicit timeouts and retry/backoff on every network fetch
  • Normalization layer that reconciles dates, addresses, and entity names into consistent fields across otherwise incompatible county formats
  • Append-only JSONL ledgers with content hashing that detect new, changed, and removed records day over day and make any historical day fully reconstructable
  • Parcel enrichment that joins records to county appraisal-district data through a shared geospatial library, with match confidence always surfaced rather than hidden
  • Scheduled daily run that emits a provenance-rich markdown brief; a human reads the watchlist and makes every origination decision

Impact

  • Fragmented public records from four counties consolidated into one consistent, current daily watchlist
  • Every line traceable to its saved source document — full provenance and audit trail by design
  • Deterministic, append-only ledgers make any prior day reconstructable for verification
  • Resilient by construction: a missing or broken source degrades a single county section, never the whole run
  • Adding a new county is configuration plus one module, with no changes to the core pipeline