Building an LLM-powered Facebook Marketplace Bot

System Overview

A bot was successfully developed to monitor Facebook Marketplace listings in Ethiopia’s emerging tech industry, leveraging a tech stack including Digital Ocean, TypeScript, and OpenAI’s gpt-4o-mini. The bot’s primary function was to filter product listings based on a query, achieving perfect results by using the LLM to differentiate between similar items like “iPhone 15” and “iPhone 15 Pro”.

Why This Matters

Ideal web scraping assumes open access, but platforms like Facebook actively employ bot detection and mitigation techniques. Circumventing these measures requires constant adaptation, and the cost of failure can include IP bans, account suspension, and wasted development effort, as demonstrated by the need for captcha harvesting and proxy rotation.

Key Insights

• Captcha Harvesting: Required to bypass Facebook login walls, using Puppeteer to locally solve captchas and transfer cookies to the VPS.
• Residential Proxies: Crucial for mimicking realistic user traffic and avoiding detection.
• Puppeteer Stealth Mode: Obscuring navigator.webdriver is vital for preventing identification as a bot.

Working Example

// Example User Agent configuration in Puppeteer
await page.setUserAgent('Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/120.0.0.0 Safari/537.36');

Practical Applications

• Use Case: Monitoring specific product availability on Facebook Marketplace, alerting users when desired items are listed.
• Pitfall: Violating platform Terms of Service (ToS) can lead to account bans and legal consequences; this implementation acknowledges violating Facebook’s ToS.

References:

• https://dev.to/isaacaddis/building-an-llm-powered-facebook-marketplace-bot-2o54