Michael F.

I build n8n and Python automations that write into Airtable, HubSpot, Close, Notion or Google Sheets, and I find the companies and decision-makers your database does not have, every row verified. English and German (native). Three things clients hire me for. ⚙️ 1 | Build: n8n automation, LLM pipelines, CRM n8n and Python flows with error handling that alerts instead of failing silently, tested before handover. LLM extraction from PDFs, emails and web pages with an only-from-the-source rule and a scrub gate for invented values. CRM sync via API: find or create contact, deal with idempotency key, owner routing, follow-up task; duplicates write nothing, broken events land in a dead-letter queue. I suggest n8n because it runs on your own server without per-operation fees; if you already use Make or Zapier, I build there. BUILT AND RUNNING: a hiring-leads tracker in n8n that pulls the open roles you care about from public job boards every day, scores each company, finds a named contact with source and confidence and upserts one row per company into Airtable; a rerun creates zero duplicates and hand-edited fields are never overwritten. A recruiting lead engine in n8n: job feeds in, company filter and LLM scoring with a verbatim quote behind every decision, decision-makers and three outreach drafts per lead written into Google Sheets. An outbound engine inside a paying client's Google Workspace (Apps Script, LLM columns per row, Smartlead API, webhook back-channel for replies and bounces), accepted in two milestones. A signal-based lead engine writing scored accounts and openers into Close CRM daily since June 2026. A monitor on my VPS that has classified 36,000+ feed items with an LLM and pushes qualified hits to Telegram within minutes. ALSO BUILT: a HubSpot intake sync (form event to contact, deal, owner, task), a trigger detector for sales-leadership changes, a job-posting monitor, quantity extraction from IFC building models. 🔎 2 | AI lead research with a data edge I do not resell database exports. I start from the sources a database cannot see: federal and state registers, licensing boards, accreditor notices, tender and permit portals, SEC filings, technographics read from a company's own job postings, partner directories. Then a verification pass on every row: source URL, date checked, email label. Catch-all domains are labelled, never sold as valid. Search stack: registers and open data first, then Exa for discovery, Parallel for per-row verification, geo-pinned Google SERP where needed; every claim is logged with source and date, negatives included, so nothing is researched twice. DELIVERED ON UPWORK: a database of every US school under one accreditor, built by joining federal education data with the accreditor's notices, 1,277 records, 283 institutions with a verified decision-maker email, delivered in three working days. 500 German construction and property groups with up to eleven decision-makers each, delivered milestone by milestone into the client's Google Sheet behind a deterministic QA gate. A technographics sample that finds companies running Snowflake or dbt from their own job postings passed an independent falsification audit. FOR AN AUSTRIAN CONSTRUCTION-DATA PUBLISHER, ongoing: CRM of 12,000 companies deduplicated and enriched against the register, a declining-activity signal feed over 806 accounts, public-tender award monitoring, pre-meeting briefings for the sales team. ALSO BUILT: a register-first universe of 3,810 US healthcare staffing agencies from 18 state registers with 305 verified decision-makers; trigger lists from ERP job postings, SEC 8-K events and Form ADV. 📬 3 | Cold email as a follow-on service When the list exists: separate sending domains, SPF, DKIM and DMARC checked from the outside, warm-up, Smartlead or Instantly through the API, verification before every push, webhooks into your sheet or CRM, weekly reporting. Your primary domain never sends a cold email. Live today for a logistics client. 🇩🇪 Deutschsprachige Projekte Prozessautomatisierung mit n8n, Make und Python, KI-Dokumentenverarbeitung (Angebote, Ausschreibungen, Pläne), Lead-Recherche aus Registern und Stellenanzeigen, Airtable, HubSpot, Google Workspace und Microsoft 365. Österreicher, Muttersprache Deutsch. Stack: Python · n8n · Apps Script · Airtable · HubSpot · Close · Notion · Clay · Smartlead · OpenAI · Gemini · Claude · Exa · Firecrawl · Docker 🤝 How I work • Access, not ownership: domains, mailboxes and platforms stay in your name; everything stays yours when we stop. • Numbers you can check: every figure above comes from an API readback or a delivered file. • Small first step: a fixed-price first milestone with written acceptance criteria, then a retainer if the system needs running. • Vienna, async, overlap with US mornings and the EU day. If this sounds like your project, write me a few lines about it. I answer within a day and I am always glad to meet new people and their problems.