Skip to content

TECHNICAL_MANUAL.md

Latest commit

 

History

History
119 lines (88 loc) · 5.23 KB

TECHNICAL_MANUAL.md

File metadata and controls

119 lines (88 loc) · 5.23 KB

💎 MRIRIDESCENT ATS RESUME MATCHER: THE MASTER TECHNICAL MANUAL

Enterprise-Grade Recruitment Intelligence & Agentic Orchestration

Developed by David Akpoviroro Oke (MrIridescent)

Digital Polymath | Purple Teamer | Systems Architect (Coding since Aug 2004)

Date: February 19, 2026 Version: 1.0.0-Gold-Production

🏛️ 1. ARCHITECTURAL PHILOSOPHY

The MrIridescent ATS Matcher is built on the principle of "Logic over Keywords." In a world where candidates use AI to "keyword-stuff" their resumes, traditional ATS systems fail. This system employs a multi-layered defense and analysis strategy to find the real talent hidden beneath the formatting.

1.1 The "Purple Team" Mindset in HR Tech

As a Cybersecurity professional (Purple Teamer), I treat the Resume as an Adversarial Input. The system is designed to:

  • Sanitize and Segment: Break down inputs into verifiable nodes.
  • Cross-Verify: Use Agentic AI to double-check claims against the raw text.
  • Forensic Scoring: Apply strict "Role-Tech-Temporal" alignment gates.

🛠️ 2. HARDWARE & SERVER SPECIFICATIONS

2.1 Developer / Entry-Level Node

  • CPU: 4-Core x86_64 (Intel Core i5 10th Gen+ or AMD Ryzen 5).
  • RAM: 8GB DDR4 (Minimum).
  • Storage: 5GB High-Speed SSD.
  • Network: Standard Broadband for API calls (Groq/Perplexity).

2.2 Enterprise Production Environment (High Concurrency)

  • CPU: 16-Core High-Frequency Compute (AWS c6g.4xlarge or equivalent).
  • RAM: 32GB - 64GB (To handle massive concurrent spaCy vector loads).
  • Storage: NVMe SSD with 100GB+ for large-scale data persistence.
  • GPU: NVIDIA A100 or RTX 4090 (If running Ollama locally for total data privacy).
  • OS: Linux (Debian 12 / Ubuntu 24.04 LTS recommended) with Docker orchestration.

🚀 3. TURNKEY DEPLOYMENT (STEP-BY-STEP)

3.1 The "One-Touch" Installation

We have automated the complexity away. Follow these three steps:

  1. Clone the Repository:

    git clone https://github.com/mriridescent/ats-matcher-pro.git
cd ats-matcher-pro

  2. Execute the Setup Wizard:

    python setup_wizard.py

    The wizard handles: Dependency resolution, NLP model hydration, .env generation, and DB schema initialization.

  3. Fire Up the Engines:

    python run.py

3.2 Environment Hardening ( .env )

Ensure your .env is configured for your desired intelligence level:

  • USE_AGENTIC_AI=true: Enables the CrewAI Multi-Agent swarm.
  • USE_OLLAMA=true: Routes all AI traffic to your local hardware (Maximum Security).
  • DATABASE_URL: Use PostgreSQL for production concurrency.

🧠 4. CORE ENGINE NUANCES

4.1 The Semantic Matching Engine (spaCy)

The system doesn't just look for "Python." It understands that "Django," "FastAPI," and "Flask" are related vectors in the Python ecosystem.

  • Model: en_core_web_md (Medium model with 20k unique vectors and 685k keys).
  • Logic: Cosine Similarity calculation between JD skill vectors and Resume experience vectors.

4.2 The CrewAI Agent Swarm

  • The JD Analyzer: Extracts structured requirements from messy text.
  • The Resume Forensicist: Breaks down the candidate's career into discrete role/tech pairings.
  • The Auditor Agent: Performs a "Hallucination Check" to ensure the AI didn't "invent" skills for the candidate.

📊 5. OPERATIONAL WORKFLOW & BEST PRACTICES

5.1 The Matching Pipeline

  1. Ingestion: Upload Job Description (PDF/Text).
  2. Structuring: AI creates a JSON representation of the role.
  3. Human Verification: The HR professional reviews and clicks Approve.
  4. Weighting: Assign "Criticality" values to specific skills (e.g., Python=100, SQL=50).
  5. Bulk Match: Upload a ZIP or multiple PDFs. The system processes them in parallel.
  6. Review Insights: View the Visual Analytics (Chart.js) to see candidate distribution.

5.2 Performance Tuning

For batches larger than 100 resumes, we recommend:

  • Increasing uvicorn worker count.
  • Using a dedicated PostgreSQL instance.
  • Setting limit_max_request_size in main.py if handling high-resolution PDF scans.

🛡️ 6. TROUBLESHOOTING & FAQ

Q: Why is the matching taking so long? A: Check if USE_AGENTIC_AI is true. Agentic flows are thorough but require more API round-trips. For speed, use basic semantic matching.

Q: How do I reset the system? A: Delete the ats_matcher.db (if using SQLite) and run python init_db.py.

📈 7. USE CASES (REAL & FICTIONAL)

7.1 The "Stealth Recruiter" (Real World)

A boutique cybersecurity firm used this system to hire an Incident Responder. The system identified a candidate who didn't use the word "Incident Response" but had extensive experience in "Digital Forensics" and "Malware Analysis"—semantic matches that traditional systems missed.

7.2 The "Startup Scale-Up" (Real World)

A fintech startup used the Bulk Upload feature to filter 400 resumes in 12 minutes, identifying the top 5 candidates with a 92% accuracy rate compared to human manual screening.

“Engineering is not just about solving problems; it is about building the tools that prevent them.” — David Akpoviroro Oke