Project Overview

JobHQ is a local-first, full-stack automation platform designed to serve the Grand Strand technical recruitment market. By orchestrating a fleet of specialized web scrapers and a tiered AI analysis layer, the system performs deep South Carolina tech market analysis, transforming raw marketplace data into high-signal recruitment opportunities for local developers and businesses.

01 // Agent-Based Scraper Architecture

The system utilizes Playwright to navigate complex, JavaScript-heavy job boards and corporate career portals.

Technical Implementation:

• Modular Scrapers: Built a decoupled "Runner" architecture where each source (Indeed, Workday, specialized medical portals) has its own navigation logic and DOM selectors.
• Browser Context Management: Implemented shared browser instances to minimize resource overhead, utilizing custom cookie persistence to bypass anti-bot challenges.
• Data Persistence: Scraped data is committed to a local SQLite database, serving as the "Raw Truth" layer before analysis.

02 // The Multi-Tier AI Analysis Engine

The core innovation of JobHQ is its Tiered AI Analysis, which balances computational cost with reasoning depth.

The Analysis Pipeline:

1. Tier 1: Pre-filtering (Local): A high-speed regex engine that instantly rejects "Trash" jobs based on strict negative keywords (e.g., "Overnight", "Part-time").
2. Tier 2: Lite Scan (Gemma 3): A lightweight AI pass using Google Gemma 3 (via OpenRouter) to perform initial sentiment analysis and basic requirement extraction.
3. Tier 3: Deep Verdict (Llama 3.3): A high-reasoning pass using Groq Llama 3.3 (70B) for top-tier candidates. This model performs deep resume-matching, extracts nuanced salary data, and generates tailored cover letters.

03 // Deterministic Scoring & Market Intelligence

Beyond simple matching, JobHQ applies a complex Scoring Weights model to rank opportunities.

• Salary Normalization: Automatically calculates hourly vs. annual wage parity and applies "Seniority Modifiers" based on local market averages.
• Penalty System: Implements deterministic score reductions for missing critical skills (-15pts), seniority mismatches (-20pts), or agency-specific listings (-10pts).
• Match Tiers: Jobs are dynamically categorized into Apply (>=85%), Possible (>=75%), or Fringe status, providing a clear daily workflow for the user.

04 // Result: Quantitative Efficiency

• Latency Reduction: Processes and scores 100+ job listings in under 5 minutes—a task that would take a human several hours.
• Data Fidelity: Extracts structured JSON data (Pay, Schedule, Requirements) from unstructured job descriptions with >92% accuracy.
• Workflow Optimization: The React-based ActivityMonitor provides real-time visibility into the AI's reasoning process, allowing for instant feedback on model decisions.

[Technical Metadata]

• Backend: Node.js (Express) / Playwright
• Frontend: React 18 / Tailwind CSS / Lucide
• AI Stack: OpenRouter (Gemma 3 / Llama 3.3 / Gemini)
• Database: SQLite