Building Job Search Copilot: A Local-First AI System for Executive Job Sourcing

The Problem: The Modern Job Search is Broken

Most job search tooling falls into two bad extremes:

1. Spammy Mass-Apply Bots: Spraying thousands of generic, low-effort resumes that get instant ATS rejections and ruin applicant reputation.
2. Manual 2-Hour Application Slogs: Re-reading JDs, manually rewriting STAR bullet points, fighting PDF margin formatting, and losing track of hiring managers in messy spreadsheets.

For senior product and engineering leaders, you need surgical precision: high-signal tailoring that maps real, verified achievements to specific JD constraints, human cover letters without AI cliches, and rapid tracking from LinkedIn sourcing to post-apply outreach.

System Architecture: 1-Line NPX & 100% Local Privacy

I designed Job Search Copilot around a strict architectural philosophy: Bring Your Own Key (BYOK) and Local-First.

When executed, the system starts a local Node.js Express server on localhost:3004 and stores all your data in ~/.job-search/data. No third-party database ever sees your private career history, visa status, hiring manager notes, or compensation benchmarks.

Core Capabilities

1. Manifest V3 Chrome Extension Sourcing

A lightweight Chrome sidebar that intercepts job postings on LinkedIn and company career boards. It cleans tracking parameters into canonical URLs, parses role requirements, and synchronizes directly with your local database in 1 click.

2. Dynamic STAR CV Compilation & Token-Optimized Generation

Instead of generic LLM rewriting that hallucinates metrics, the engine uses your verified career history as a strict source of truth. It selects the highest-relevance 4–5 roles, reframes bullets using the STAR (Situation-Task-Action-Result) methodology, and produces pixel-perfect 2-page PDFs.

Using Gemini 2.5 Flash, each full CV tailoring and PDF compilation costs less than $0.0001 (a hundredth of a cent).

3. Rejection Cohort Intelligence & A/B Simulation

When applications are dismissed or rejected, the system runs aggregate cross-role failure analysis across the cohort. It identifies whether rejections were due to domain mismatches, framing gaps, or under-emphasized skills, and suggests prompt adjustments.

Before applying new rules globally, you can run a Side-by-Side A/B Simulation to see how Version A vs Version B transforms your CV headlines and bullet highlights on real job descriptions, complete with live token impact metrics.

Try It Locally

The project is fully open source. To run it immediately with zero installation overhead:

npx job-copilot

Clone the repository, contribute, or inspect the code on GitHub.