JobForge AI

Autonomous multi-agent job hunting pipeline — discovers, evaluates, and prepares personalised applications for Data Science, AI Engineering, and ML Engineering roles across the UK.

Built with LangGraph for stateful agent orchestration, Google Gemini for LLM-powered reasoning, and a modular connector architecture.

What It Does

Every morning, JobForge AI:

1. Scouts 6+ job sources (Adzuna, Reed, Wellfound, LinkedIn, Indeed, startup career pages) for fresh AI/ML/DS roles
2. Scores each job against your CV using a dual-pass engine (embedding pre-screen + 6-dimension LLM evaluation)
3. Tailors a job-specific PDF CV for every qualified job (≥70% match), selecting the optimal CV variant
4. Dispatches a curated Excel digest + tailored PDFs to your inbox

It never auto-applies. The system stops at generating tailored CVs and an email digest.

Key Features

• Deep Agent Architecture: Scout, Matchmaker, and Tailor agents use hierarchical planning, persistent memory, and self-reflective evaluation loops
• Visa Intelligence: Detects sponsorship signals in job descriptions. PSW-aware scoring — all roles are valid, sponsoring roles get a strategic boost
• Startup Focus: Dedicated Wellfound/AngelList connector + curated UK AI startup watchlist
• Zero Hallucination CV Tailoring: Skill Inventory ground truth ensures the Tailor Agent never invents skills or metrics
• Cross-Run Deduplication: SQLite-backed hash store ensures the same job is never re-processed

Quick Start

# Clone
git clone https://github.com/Viraj97-SL/jobforge-ai.git
cd jobforge-ai

# Setup
python -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"
cp .env.example .env  # Fill in your API keys

# Extract skill inventory (run once)
python scripts/extract_skill_inventory.py

# Test the Scout Agent
python scripts/run_pipeline.py --scout-only

# Run full pipeline
python scripts/run_pipeline.py

Tech Stack

Layer	Technology
Agent Orchestration	LangGraph 0.2.x + LangChain 0.3.x
LLM	Google Gemini (Flash + Pro)
Job Sources	Adzuna, Reed, Wellfound, Tavily, SerpAPI
Persistence	SQLite (local) / PostgreSQL (prod)
CV Compilation	pdflatex (TeX Live)
Excel	Pandas + openpyxl
Email	SMTP / Resend
Testing	Pytest + pytest-cov
CI/CD	GitHub Actions

Architecture

[CRON] → Scout Agent → Matchmaker Agent → Tailor Agent → Dispatcher Agent → [EMAIL]
              │                │                │                │
         Fan-out search   Dual-pass score   CV variant select   Excel + PDF
         6+ sources       6 dimensions      LaTeX modify        Email dispatch
         Dedup store      Visa adjustments  Hallucination check

Visa Context

The system is designed for a Graduate Route (PSW) visa holder:

• Full work rights for 2 years — can accept any role
• Sponsoring roles get a +10 point scoring boost (strategic long-term value)
• "UK citizens only" roles get a -5 point penalty but are NOT excluded
• Jobs are tagged with visa status in the Excel digest for informed decision-making

License

MIT

---

Built by Viraj Bulugahapitiya | MSc Data Science | AI Engineer