Matias K.

Most of the data that would actually move your business forward isn't sitting in a clean database waiting for you. It's locked inside scanned PDFs, hidden behind CAPTCHAs, buried in government portals with no API, or scattered across platforms that actively don't want to be scraped. That's where I come in. Python, Playwright and FastAPI on GCP — hard-target scrapers, OCR/LLM document pipelines, and production ETL that runs without babysitting. 🏆 Top Rated · 100% Job Success · 20+ projects delivered across legal, real estate, SaaS and agency clients ━━━━━━━━━━━━━━━━━━━━━━ THE FAILURE MODE I'M SET UP TO PREVENT Most scraping and extraction projects don't fail on day one. They fail in week three. The site changes its layout, a session expires, a 2FA prompt appears, the anti-bot vendor pushes an update — and the script dies quietly while your team keeps trusting the data. I build for week three, not for the demo: persistent sessions, per-source error recovery, structural drift detection, and alerts that fire before the pipeline goes silent. ━━━━━━━━━━━━━━━━━━━━━━ RECENT PRODUCTION RESULTS ⚖️ Forensic title research — feasibility audit for a real estate investor targeting North Carolina court records. I ran a reconnaissance agent against the live TylerTech eCourts portal and a legacy county Register of Deeds system, mapped their exact defenses (reCAPTCHA, JavaScript session heartbeat, anti-forgery tokens, a PHPSESSID disclaimer gateway), and delivered the pipeline architecture and NLP signal set — before a line of production code was written. 🌍 Nationwide directory extraction — 100K+ structured business records behind session auth and anti-bot protection, delivered incrementally. Client feedback: "Exceptional communication and really high quality work." 🤖 Multi-source content + OSINT engine — ingestion from Meta, X and TikTok APIs into a Claude-powered generation pipeline. Now running in production and expanded by the client into a white-label product. 🔁 24/7 browser automation on GCP — replaced a paid no-code tool that broke on every UI change with a Playwright pipeline running unattended on Cloud, surviving 2FA re-verification and a full SPA redesign. ━━━━━━━━━━━━━━━━━━━━━━ WHAT I BUILD ✔️ Hard-target scraping — Playwright stealth, proxy rotation, fingerprint management, session persistence; Cloudflare and CAPTCHA-gated portals, government registries with no API. ✔️ Document AI pipelines — OCR (Tesseract, Vision AI) plus LLM classification over scanned PDFs, court filings and government records. Pydantic-validated schemas, so no field is ever hallucinated. ✔️ ETL & cloud infrastructure — FastAPI, Pandas, SQLAlchemy on GCP Cloud Run, landing in PostgreSQL or BigQuery. Incremental, fault-tolerant, scheduled. ✔️ AI integration — Claude API with Batch and prompt caching for real cost control, Claude Code workflows, RAG over extracted documents, structured JSON extraction. ✔️ Workflow automation — Make/n8n orchestration, WhatsApp Business API, CRM integrations, scheduled pipelines. ━━━━━━━━━━━━━━━━━━━━━━ HOW I WORK ✅ I scope before I code — sources reviewed, edge cases mapped, milestone plan delivered. You know what you're getting and when, before you commit to anything. ✅ I own the system end to end: architecture → build → deploy → monitoring → handover docs. ✅ I'll tell you when scraping is the wrong answer — and when a $40/month API beats a $3,000 pipeline. ✅ Async-first: Loom walkthroughs and written updates at every milestone. Available for a call when a decision genuinely needs one. ━━━━━━━━━━━━━━━━━━━━━━ ONGOING WORK Data pipelines are not a one-time build. Sites change, formats drift, volumes grow. I take on a small number of ongoing engagements — weekly retainer or embedded hours — for teams that need someone to own their extraction layer instead of receiving a script and a goodbye. Currently open to two of these. ━━━━━━━━━━━━━━━━━━━━━━ TO GET STARTED Send me: - Target site(s), or a sample document - The fields you need - Expected volume - Update frequency - Preferred output (CSV, Postgres, BigQuery, API) I'll come back within 24 hours with a technical assessment and a fixed-price milestone plan — not a sales pitch. If it isn't a fit, I'll say so and point you to what is. keywords: python, playwright, selenium, scrapy, fastapi, web scraping, data scraping, web crawling, data extraction, data mining, ocr, tesseract, document ai, pdf extraction, court records, government portals, public records, title search, legal data, etl, data pipeline, bigquery, postgresql, gcp, anti-bot, cloudflare, captcha, proxy rotation, claude, claude api, claude code, openai, llm, rag, structured extraction, whatsapp api, make automation, n8n, real estate data, directory scraping, lead generation, b2b data, price monitoring, competitive intelligence, apify, supabase