JobPilot

A personal job application assistant that crawls job postings, analyzes CV fit using LLMs, and tracks the entire application process — all through a self-hosted web dashboard.

---

What it does

Job hunting generates a lot of noise. This tool cuts through it:

1. Crawls Indeed job postings via RSS feeds and a Playwright scraper
2. Analyzes each posting against your CV using an LLM — producing a fit score (0–100 %), level assessment, strengths/gaps, and concrete CV improvement suggestions
3. Tracks the full application lifecycle: applied, response received, interview, rejection
4. Maintains a candidate profile — one-time CV analysis produces a compact profile (~300 tokens) that replaces the raw PDF in every subsequent analysis (~75 % token reduction)
5. Integrates GitHub — scans all public repositories, extracts tools and skills from READMEs, and merges them into the candidate profile

---

Tech stack

Layer	Technology
Backend / CLI	Python 3.12, asyncio
Web UI	Streamlit
Database	MongoDB (pymongo)
LLM integration	Anthropic Claude API (claude-opus-4-6) · OpenAI-compatible (LM Studio)
Web scraping	Playwright (Chromium), feedparser, httpx
PDF parsing	pypdf
Containerization	Docker, Docker Compose
Scheduling	APScheduler
Notifications	plyer (desktop), SMTP (email digest)

---

Architecture

┌─────────────────────────────────────────────────────┐
│                  Streamlit Dashboard                 │
│  Overview · Job List · Analysis · Profile Manager   │
└────────────────────┬────────────────────────────────┘
                     │
        ┌────────────▼────────────┐
        │        MongoDB          │
        │  jobs · search_runs     │
        └──────┬──────────────────┘
               │
    ┌──────────┴──────────┐
    │                     │
┌───▼────────┐    ┌───────▼──────────────────┐
│  Fetcher   │    │      LLM Analyzer         │
│  RSS Feed  │    │  Anthropic Claude API     │
│  Playwright│    │  or local LM Studio       │
└────────────┘    │  (OpenAI-compatible)      │
                  └───────────────────────────┘

Key design decisions:

• Dual LLM backend — switches between cloud (Anthropic) and local (LM Studio) via a single config flag; falls back automatically if the local model is unreachable
• Candidate profile pattern — CV is analyzed once; the resulting Markdown profile replaces the raw PDF in all subsequent LLM calls, drastically reducing latency and cost
• Schema-less extra fields — core job data uses a typed dataclass; analysis results (full text, model name, token counts, levels) are stored as extra MongoDB fields without schema migration
• Stateless dashboard — all state lives in MongoDB and st.session_state; the Streamlit process can be restarted at any time without data loss

---

Features

Analysis

• Fit score with reasoning (0–100 %)
• Candidate level vs. required level assessment (Junior / Mid / Senior / Lead)
• Strengths and structured gap analysis per job posting:
  • Missing tools & technologies (bulleted list)
  • Missing experience & knowledge areas (bulleted list)
  • Other gaps: soft skills, certificates, languages, domain experience (bulleted list)
• Concrete CV improvement suggestions per gap
• Extracts job title and company automatically from the posting text

Candidate profile

• One-time LLM analysis of CV + personal notes (me.md)
• Granular skill extraction: programming languages, ML/AI frameworks, databases, tools
• GitHub integration: fetches READMEs from all public repos, extracts technologies via LLM
• Editable in the dashboard; profile is used automatically in all analyses

Application tracking

• Status pipeline: new → applied → interview → offer / rejected / withdrawn
• Tracks application date, response date, interview invitations, rejection notes
• Statistics overview: scores, response rates, status distribution

Infrastructure

• Fully containerized with Docker Compose (app + MongoDB)
• Configurable crawl schedule (time window, weekdays only, interval)
• Rate limiting on scraper requests (randomized delays)

---

Getting started

Requirements: Docker, or Python 3.12 + MongoDB

# Clone and configure
cp .env.example .env
# Edit .env: set ANTHROPIC_API_KEY or point LM_STUDIO_URL to your local instance

cp config/settings.yaml.example config/settings.yaml
# Edit config/settings.yaml: set lmstudio_url and cv.path

# Start everything
docker compose up -d

# Dashboard
open http://localhost:8501

Local development:

python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
playwright install chromium

MONGODB_URI=mongodb://localhost:27017/jobtracker \
  python -m streamlit run dashboard.py

---

Configuration

config/settings.yaml:

analyzer:
  backend: "lmstudio"          # "anthropic" or "lmstudio"
  lmstudio_url: "http://192.168.1.x:1234/v1"

cv:
  path: "data/cv.pdf"
  profile_path: "data/kandidatenprofil.md"   # auto-generated

scheduler:
  crawl_interval_hours: 2
  crawl_window_start: 8
  crawl_window_end: 18
  crawl_weekdays_only: true

---

Project structure

├── src/
│   ├── analyzer/job_matcher.py    # LLM analysis, profile creation, CV improvement
│   ├── dashboard/streamlit_app.py # Full Streamlit UI
│   ├── fetcher/                   # RSS + Playwright scrapers
│   ├── storage/                   # MongoDB models and connection
│   ├── notifier/                  # Desktop + email notifications
│   └── scheduler/                 # APScheduler jobs
├── config/
│   ├── settings.yaml
│   └── search_profiles.yaml
├── main.py                        # CLI entrypoint
├── dashboard.py                   # Streamlit entrypoint
├── Dockerfile
└── docker-compose.yml