JobHunter

AI-powered job application pipeline. Scrapes 100+ job boards, scores jobs, researches companies, generates personalized outreach emails, and sends them — all at $0 in LLM costs via free-tier providers.

# Quick start (download tarball)
curl -fsSL https://github.com/arinbalyan/jobhunter/releases/latest/download/jobhunter-x86_64-linux.tar.gz \
  | tar xz
cd jobhunter-*
cp config.example.toml config.toml   # edit with your details
./jobhunter doctor                    # verify setup

# Or build from source
git clone https://github.com/arinbalyan/jobhunter.git
cd jobhunter
cargo build --release && cd scraper && go build -o . . && cd ..
cp config.example.toml config.toml
./target/release/jobhunter doctor

---

Architecture

jobhunter (Rust binary, ~700 lines)
  │
  ├── spawns → scraper (Go bridge, ~30 lines)
  │               │
  │               └── scrappy v0.3.9 (Go library: 141 sites, email extraction, MX verify)
  │
  ├── LLM router (9 free providers, weighted random + failover)
  ├── SMTP sender (Gmail 587 STARTTLS, rate-limited, quota tracked)
  ├── Tracking server (open/click pixels)
  ├── Pipeline dashboard (Vercel)
  └── Telegram alerts (per-run reports)

Two binaries in one tarball. The Go scraper is a thin bridge to scrappy.

Commands

Command	What it does
jobhunter scrape --mode remote|onsite	Scrape 141 boards → filter → dedup → queue
jobhunter score	Score unscored jobs 1-10 via LLM
jobhunter research	Generate 3 talking points per company via LLM
jobhunter send	Generate emails + send via SMTP
jobhunter triage "<reply>"	Classify recruiter reply (positive/negative/neutral)
jobhunter import --from <scrappy_config>	Import scrappy per-site config into JobHunter config
jobhunter serve	HTTP tracking server + pipeline dashboard
jobhunter doctor	Diagnose config, DB, API keys, binary

Pipeline

Scrape → Score → Research → Send → Track
  │        │        │         │        │
  │        │        │         │        └── opens (1x1 pixel)
  │        │        │         │        └── clicks (/click?e=&url=)
  │        │        │         └── Gmail SMTP (rate-limited 1/15s)
  │        │        │             └── signature with GitHub/Portfolio/Resume
  │        │        └── 3 talking points per company (LLM, simple model)
  │        └── 1-10 match score per job (LLM, simple model)
  └── 141 boards → title filter → email filter → SQL dedup → email_queue
      scrappy guest API (no Playwright needed)

Configuration

Single file .data/config.toml. API keys via $VAR references from .env or GitHub Secrets.

Key sections:

Section	Purpose
[user]	Name, role, experience, GitHub, portfolio, resume URL, LLM context
[search.remote] / [search.onsite]	Search terms, locations, sites per mode
[scrape]	Runtime, results limit, reject_titles, blocked_email_* (all configurable)
scrappy_config	Path to scrappy's config.toml — auto-loads 118 per-site search overrides
[sites.*]	Per-site search terms/location overrides (optional, auto-imported)
[[llm.providers]]	8 free providers with weights, models, API key env vars
[templates.*]	All LLM prompts — edit without recompiling

Full reference at docs/configuration.md.

LLM Providers (All Free Tier)

Provider	Complex Model	Weight
OpenRouter	google/gemma-4-31b-it:free	10
Groq	openai/gpt-oss-120b	5
Together	google/gemma-4-9b-it	4
DeepInfra	Llama-3.3-70B-Instruct-Turbo	4
Hyperbolic	qwen3-coder-480b-a35b-instruct:free	3
SambaNova	Meta-Llama-3.3-70B-Instruct	3
Cerebras	zai-glm-4.7	2
Z.AI	GLM-4-Plus	1

Weighted random selection, failover up to 3 providers, auto-cooldown on failures.

LLM Features

Feature	Model	Cost
Email generation	Complex (e.g. Gemma 4 31B)	One call per email
Job scoring (1-10)	Simple	One short call per job
Company research (3 points)	Simple	One call per company
Reply triage	Simple	One call per reply

Deployment

GitHub Actions (automated)

Three workflows:

• scrape — 4x daily, builds Rust + Go bridge, scrapes all 78 sites
• send — daily, generates + sends emails for queued jobs
• tests — on every push, runs cargo test + Go vet

Vercel Dashboard

Live at https://jobhunter-tracker.vercel.app — pipeline funnel, per-URL click breakdown, run history, failures. Node.js serverless functions in api/.

Self-hosted tracking server

./jobhunter serve
# → GET /         pipeline dashboard (HTML)
# → GET /track?e= 1x1 GIF, logs open
# → GET /click?e=&url= 302 redirect, logs click
# → GET /health

License

MIT