Joel Martinez Seattle

The Workshop · 11

Signal-Index Engine

A propensity engine that scores thousands of brands on composite health signals, paired with a crash-safe enrichment pipeline that picks up exactly where it died.

Python Systems architecture Resilient ETL
3,400+
decision-maker contacts verified across runs
~2,800
brands per sourced universe (industry × headcount × geo)
0
records lost to a crash — the pipeline resumes from where it stopped

The problem

A client needed to score thousands of consumer brands on composite "health" signals — repeatably, in a way that could be re-pointed at future clients — and to source and verify decision-makers at scale without losing hours of work every time a long enrichment run crashed.

What I built

A reusable orchestrator with per-signal data-agent modules on a shared interface, driven by a per-tenant config file, feeding a leaderboard. Alongside it, an enrichment pipeline built to survive the real world: retries on connection resets, incremental write-and-flush so partial progress is always saved, per-record isolation so one bad row can't sink the batch, and resume-from-file so a restart skips everything already processed.

The result

  • 3,400+ decision-maker contacts verified across live runs.
  • Brand universes of roughly 2,800 sourced by industry, employee band, and geography.
  • Architected for multiple tenants on one common interface (one instantiated to date).