Job Pipeline | Project Memo

Shipped proof

Active, operator-run daily

01
Live funnel from a 1,228-lead triage queue: 92 leads promoted, 26 opportunities applied, 25 ready to submit, 21 researching, 1 interviewing.
02
Release wave priority-remote-2026-06-06: 16 tailored apply packets generated in four batches, each a styled resume, an ATS variant, and a cover letter, with fit 62–87 and resume audit 88–94.
03
Outcome learning loop over 46 applications surfacing notion_seed outbound (20% response, 80% progression) as the top channel while flagging bulk-apply paths that returned zero responses.

PlannedPipeline architecture

Bounded claim

An AI job-search pipeline built so it can't lie: every generated résumé is checked against a verified fact base before it ships, and a human sends every application. 1,228 leads triaged, 16 audited apply packets.

RoleFounder / builder

StatusActive, operator-run daily

Available proof0 / 2

Sourceoperator-recorded

Stack

Project

Fact-safe job-search pipeline

A second domain on the same Symphony orchestration substrate: ingest, triage, research, fact-safe auto-draft, independent audit pass, apply-packet release, and a truthful outcome learning loop the operator runs daily. V3 ships a structural defense against hallucinating AI-apply tools: tailored résumé content is validated against the base-resume corpus before any PDF renders, making fabrication architecturally impossible.

Problem

Operating context

A live job search has two compounding failure modes: volume drowns judgment (thousands of raw leads, manual resume tailoring, no channel signal), and AI-apply tooling introduces a new risk — fabricated metrics and inflated claims that ship in generated resumes because no structural gate exists. Getting the funnel fast enough to run daily also requires fixing the measurement layer so parked drafts and never-sent covers do not silently inflate conversion stats.

Ownership

What I owned

Built and runs the V3 pipeline on the Symphony orchestration stack — fact-safety gate, lead ingest and triage scoring, auto-drafter, independent audit pass, batched apply-packet generation, and a truthful outcome learning loop — operating the operator's own search through it daily.