RJ DowellPERSONAL PORTFOLIO
← All projects

PROFESSIONAL WORK · Recruiting data & AI

Candidate matching prototype

A read-only prototype that rolls structured notes up into candidate review packets for recruiters.

BuiltRead-only prototypeDocumented September 25, 2026

My contribution

Developed the extraction, quality checks, candidate rollups, and matching preview for a read-only proof of concept.

Tools
Python · Bullhorn · Structured extraction · Deterministic eligibility checks

How it works

  1. Read eligible notes and filter irrelevant content before model extraction.
  2. Validate structured output and generate an idempotent preview.
  3. Roll up candidate evidence and compare it with role requirements.
  4. Prepare a packet for recruiter review.

Decisions that mattered

A license match is only one dimension

Readiness also depends on interest, availability, and other explicit constraints. Retirement, opt-out, and incompatible work preferences block a misleading recommendation.

Selection order can bias the result

An early low-ID selection pattern favored older records. The investigation treated data selection as part of model quality.

Keep the experiment reversible

The prototype doesn’t write to the CRM, insert production records, or post packets to a team channel.

Results

The September 25, 2026 record documents the read-only workflow and a 674-test suite.

Scope & limits

Built to support recruiter review; its effect on placements hasn’t been measured.

Related projects

Next projectScheduling data workspace