fb-rental-scanner

Automates apartment hunting by scraping Facebook groups, extracting structured data with an LLM, and generating ready-to-send messages to landlords.

Because scrolling Facebook groups for hours is boring.

How it works

1. Scrape — opens Facebook groups in Chrome, scrolls to the last unseen post, and collects new posts (author, text, timestamp, URL)
2. Parse — sends each post to an LLM to extract structured data: price, location, property type, utilities, move-in date, etc.
3. Filter — applies your criteria from user_config.yaml (max price, location, property type, etc.)
4. Review — manually go through filtered apartments and keep/remove them
5. Message — generates ready-to-send messages for each remaining apartment

---

Usage

Automated mode

make monitor-apartments

Runs scraper and parser in a loop every N minutes (configurable in system_config.yaml). Prompts for manual review after each cycle if new relevant apartments are found.

Step-by-step (Jupyter)

Open demo.ipynb and run cells one by one:

1. Login — opens Chrome, auto-fills credentials if provided, waits for CAPTCHA
2. Scrape — collects new posts from all configured groups
3. Parse — extracts structured data and filters by criteria
4. Manual operations — review, add, or generate messages

---

Quick setup

cp .env.example .env
cp user_config.example.yaml user_config.yaml

1. Add FB credentials + Groq API key (or use local Ollama) in .env
2. Edit user_config.yaml with your groups, message template and criteria
3. Optionally edit system_config.yaml to change LLM, scrape interval, and timing

---

CLI commands

make monitor-apartments  # run scraper + parser in a loop
make filter-raw          # reprocess raw scraped files
make add-apartments      # manually add apartments
make review-apartments   # review and remove apartments
make generate-messages   # generate messages for all relevant apartments

---

Project structure

fb-rental-scanner/
├── demo.ipynb                     # step-by-step Jupyter walkthrough
├── user_config.yaml                # groups, criteria, message template
├── system_config.yaml              # LLM, scraper, and runner settings
├── .env                            # secrets
├── scr/
│   ├── runner.py                   # main loop: scrape → parse → review
│   ├── scraper.py                  # selenium scraping logic
│   ├── parser.py                   # LLM extraction and filtering
│   ├── models.py                   # pydantic models
│   ├── setup.py                    # config loading and logging setup
│   └── manual_operations.py        # manual review, add, and message generation
├── data/
│   ├── raw_data/                   # raw scraped CSVs (auto-deleted after parsing)
│   ├── unsorted_apartments.csv     # all extracted apartments before filtering
│   ├── relevant_apartments.csv     # filtered apartments
│   └── messages_<timestamp>.txt    # generated messages ready to send
└── makefile

---

Configuration

user_config.yaml — things you change often:

• facebook_groups — list of group URLs to scrape
• criteria — price range, property type, location, etc.
• message_template — message sent to landlords

system_config.yaml — infrastructure settings:

• llm_config — model type (groq or local), model name, prompt
• scraper_config — cooldown, scroll speed, CSS/XPath selectors
• runner_config — scrape interval in minutes

---

LLM support

Supports any model via LangChain. Tested with:

• Groq (meta-llama/llama-4-scout-17b-16e-instruct) — fast, free tier available
• Ollama (llama3) — fully local, no API key needed

Change model_type in system_config.yaml to switch.

---

Notes

• Facebook requires manual CAPTCHA solving on first login
• Images and videos are blocked after login to speed up scraping
• last_visited per group is saved automatically in user_config.yaml after each run
• LLM uses structured output — null values and type coercion are handled automatically

---

Disclaimer

This project was built for personal and educational purposes only. Scraping Facebook may violate their Terms of Service. The author is not responsible for any misuse or consequences of using this tool. Use at your own risk.