Italy Job Hunter

An AI pipeline that automates the entire job hunting workflow for the Italian tech market — from web search to personalized CV analysis — and delivers results directly to Telegram.

Built because manually scanning job boards every day is a solved problem. This tool runs unattended, filters noise with a fast LLM, and only surfaces opportunities that actually match your profile.

---

How It Works

Two independent pipelines, both running fully automated:

Hunt mode — finds active job listings targeting the Italian market, runs a fast boolean AI triage, analyzes each match against your CV, and delivers a formatted report to Telegram.

Scout mode — discovers target companies in Italy and generates a personalized cold-outreach pitch for each one, ready to send to a CTO or HR Manager.

Tavily (web search)
       │
       ▼
Groq / llama-3.1-8b-instant
(boolean triage — YES / NO)
       │ YES
       ▼
DeepSeek-V3
(CV match analysis + recruiter hook)
       │
       ▼
Telegram Bot
(formatted report)

Each model is assigned the task it handles best: Groq for fast, cheap boolean classification — DeepSeek for deep reasoning and personalization. No single model does everything.

---

Why Multi-Model

A single LLM for the entire pipeline is expensive and slow. This architecture uses:

• Groq/llama-3.1-8b-instant for triage — inference in milliseconds, near-zero cost, perfect for a binary yes/no filter
• DeepSeek-V3 for analysis — high reasoning quality at a fraction of GPT-4 pricing, handles nuanced CV matching

The result: processing 20 job listings costs less than a cent and completes in under 30 seconds.

---

Prerequisites

• Node.js 18 or higher
• API keys for the following services (all have free or very cheap tiers):

Service	Purpose	Cost
Tavily	Web search	Free tier available
Groq	Boolean triage	Free
DeepSeek	CV analysis	Pay-as-you-go, very cheap
Telegram Bot	Delivery	Free

---

Installation

# 1. Clone the repository
git clone https://github.com/simonecamerano/italy-job-hunter.git
cd italy-job-hunter

# 2. Install dependencies
npm install

# 3. Set up environment variables
cp .env.example .env
# Edit .env and fill in your API keys

# 4. Add your CV
# Replace data/cv.md with your own CV in Markdown format

---

Usage

# Find job listings and send match reports to Telegram
npm start

# Find target companies and generate cold-outreach pitches
npm run scout

# Run the test suite
npm test

---

Configuration

Environment variables (.env)

Variable	Description
TELEGRAM_BOT_TOKEN	Your bot token from @BotFather
TELEGRAM_CHAT_ID	Your personal chat ID (use @userinfobot)
TAVILY_API_KEY	Tavily API key
GROQ_API_KEY	Groq API key
DEEPSEEK_API_KEY	DeepSeek API key

Search constants (src/config.js)

All search queries, model names, and tuning parameters live in src/config.js. Edit this file to adapt the tool to any stack or market without touching the pipeline logic.

Constant	Default	Description
SEARCH_QUERY	(see file)	Tavily query for job listings
SEARCH_MAX_RESULTS	20	Max raw results per hunt run
SCOUT_QUERY	(see file)	Tavily query for company scouting
SCOUT_MAX_RESULTS	6	Max companies per scouting session
TRIAGE_MODEL	llama-3.1-8b-instant	Groq model for boolean filtering
ANALYSIS_MODEL	deepseek-chat	DeepSeek model for CV analysis
API_DELAY_MS	2500	Delay between calls (rate limit safety)
TELEGRAM_MAX_CHARS	4000	Message chunk size (Telegram limit: 4096)

CV (data/cv.md)

Replace data/cv.md with your own CV in Markdown format. The analyzer reads it at runtime to generate personalized match reports and cold-outreach pitches. The quality of the output scales directly with the quality of your CV file.

---

Project Structure

italy-job-hunter/
├── data/
│   └── cv.md                     Your CV in Markdown format
├── src/
│   ├── config.js                 Centralized constants — queries, models, limits
│   ├── seen_store.js             Deduplication cache — no duplicate results
│   ├── search_engine.js          Stage 1: web search via Tavily
│   ├── triage_filter.js          Stage 2: boolean AI filter via Groq
│   ├── deepseek_analyzer.js      Stage 3: CV match analysis via DeepSeek
│   ├── telegram_sender.js        Stage 4: formatted delivery via Telegram
│   ├── company_scouter.js        Scout: finds target companies via Tavily
│   └── spontaneous_analyzer.js   Scout: generates cold-outreach pitches via DeepSeek
├── tests/                        Vitest test suite — runs fully offline, no API calls
├── index.js                      Hunt mode orchestrator
├── scouting.js                   Scout mode orchestrator
└── .env.example                  Environment variable template

---

Tech Stack

Tool	Purpose	Why
Tavily	Web search	Purpose-built for AI agents; returns clean, structured content
Groq	Triage filter	Fastest inference available — ideal for high-volume boolean classification
DeepSeek-V3	CV analysis	High reasoning quality at a fraction of GPT-4 pricing
Telegram Bot API	Delivery	Instant push notifications, zero UI to build or maintain
Vitest	Testing	Native ESM support, zero config

---

Adapting to Your Stack

The pipeline is not Italy-specific. To use it for a different market or tech stack:

1. Edit SEARCH_QUERY and SCOUT_QUERY in src/config.js
2. Replace data/cv.md with your own CV
3. Swap model names if you prefer different providers — the interfaces are modular

The core pipeline logic does not need to change.

---

Author

Simone Camerano — AI workflow engineer and full stack developer.

I build systems that automate complex, multi-step processes. This tool came out of a real problem: the Italian tech job market is fragmented and noisy. Manual searching wastes time that could be automated.

• 🌐 simonecamerano.dev
• 💼 linkedin.com/in/simone-camerano
• 🐙 github.com/simonecamerano

---

License

ISC