Automated Web Scrapers & Data Pipelines

**DealHunter** aggregates used production equipment listings from 8 marketplaces (Craigslist, eBay, Facebook, Mercari, KEH, MPB, Reddit, OfferUp) and job postings from 13+ boards, surfacing deals via Telegram alerts. Built with Flask, APScheduler, and BeautifulSoup—scrapers run every 20 minutes for gear, every 2 hours for gigs—storing results in SQLite with WAL mode for concurrent access. Local dashboard at port 9700 filters by price, condition, and location across all sources simultaneously.

DealHunterSA aggregates tax sales, struck-off properties, and foreclosure data from Bexar County public sources into a real estate investment intelligence platform, automating deal discovery for distressed-property investors. Built with Python scrapers (Playwright, pdfplumber, pytesseract), SQLite persistence, and a Flask dashboard, it scores leads across multiple distress signals and generates direct-mail campaigns. Live at https://dealhunter.thirstyai.live with 5 core pipeline tools handling PDF monitoring, investment analysis, and automated lead qualification.

GVTriage automates webpage classification and triage using Claude's API to intelligently parse and categorize web content at scale. Built with BeautifulSoup for HTML extraction and Anthropic's models for semantic understanding, it eliminates manual content sorting workflows. Deployed as a serverless pipeline processing hundreds of pages daily with minimal latency overhead.

HOALeads automatically extracts homeowners association contact data from public property records and websites, eliminating manual research for real estate professionals. Uses Python requests library to scrape and parse structured HOA databases, aggregating results into exportable lists. Processes 500+ property records per run with 95%+ contact accuracy.

**IGScout** scrapes Instagram location pages on-demand via Playwright to surface sales prospects, scores posts with Claude Haiku, and drafts contextual replies—eliminating manual scrolling for location-based prospecting. Flask dashboard triggers headless scrapes of 5 San Antonio locations, stores results in SQLite, and routes outputs to Telegram; deployed live behind Cloudflare tunnel with session persistence via cookie injection. The system prioritizes account safety over scale, favoring deliberate on-demand scans over risky background automation.

**M3Hunter** — Real-time deal aggregator scraping four major auto marketplaces (auto.dev, Cars.com, Craigslist, BaT) to surface BMW M3 Competition listings, tracking price drops and alerting via iMessage/Telegram. Uses httpx + BeautifulSoup for multi-source scraping, SQLite for price history and deduplication, and a Rich CLI dashboard alongside a Flask web interface with scheduled background jobs. Deployed to Render with configurable scrape intervals and currently monitoring active market listings at scale.

MacHunter aggregates used/refurbished Mac listings from eight sources (Craigslist, eBay, Apple Refurbished, Google Shopping, Facebook, Swappa, BackMarket) into a unified dashboard, solving price-monitoring friction for deal hunters in San Antonio. Built with Flask + SQLite + APScheduler, it scrapes every 15 minutes with source-specific BeautifulSoup parsers, persists listings in a local database, and batches alerts via Telegram. Serves a live dashboard on localhost:9860 with price filtering and listing dismissal.

**RetailScout** identifies high-intent retail leads by scanning nearby storefronts and instantly rating their digital presence—no website = hot prospect, weak site = warm, solid site = cold. A Cloudflare Worker proxies Google Places API with 7-day grid-cell caching, then parallel HEAD-checks each shop's SSL and mobile responsiveness. Deployed as a Safari-installable PWA with IndexedDB persistence, live at scout.thirstyai.live.

**UpworkHunter** automatically discovers and evaluates Upwork jobs matching AI/automation keywords, scoring each with DeepSeek AI ($0.01/job cost) against buildability and feasibility thresholds, then spawns Claude Code agents to generate working prototypes—all orchestrated via CLI with concurrent build limits, SQLite persistence, and Telegram notifications.