Systems that work while you think.

Stop paying humans to copy-paste data from Apollo. This system automates the entire top-of-funnel process—from raw lead capture to automated outreach—while keeping a human in the loop to guarantee quality control.

The Pipeline:

1. Automated Sourcing: The system pulls leads directly from the Apollo API based on your exact ICP criteria and logs them into a lightweight custom CRM.
2. AI Pre-Qualification: It scrapes the lead’s website and uses AI to verify they actually have an active blog. No blog? Disqualified instantly. Zero human time wasted on dead ends.
3. Human-in-the-Loop (HITL) Verification: Pre-qualified leads are pushed to a custom Telegram bot in batches of 10. A human does a 10-second visual check and hits “Qualify” or “Disqualify.”
4. Automated Enrichment: Once approved, the system automatically pulls C-Suite and executive contacts for that company and adds them to the CRM.
5. One-Click Outreach: The human reviews the contacts via Telegram, selects the target, and the system automatically kicks off a customized email outreach sequence.

Why Agencies Should Care:

This isn’t a fragile script; it’s a fully functioning lead generation department that never sleeps. You can sell this to your clients as a “Done-For-You Lead Gen” service, charge $5k+ for the setup, and your only ongoing cost is having a VA do 10-second Telegram approvals. It combines n8n, Apollo, AI, and Telegram into one cohesive, revenue-generating system.

Stop paying developers to push simple content updates. This site is a zero-friction, hot-reloading content engine built on Eleventy and Nunjucks, hosted on Dockerized Hetzner infrastructure.

The Pipeline:

1. Obsidian-Native Authoring: All copy, metadata, layouts, and navigation are authored entirely in plain text Obsidian frontmatter. No fighting with clunky CMS interfaces or WordPress bloat.
2. Instant Sync & Build: Content syncs instantly from Obsidian to the sandbox. Eleventy builds and hot-reloads immediately. Zero git push required.
3. Agent-Managed Deployment: The site is wired directly into the Decapod agentic framework. AI agents can update content, proofread, fact-check, and deploy changes seamlessly without a human ever touching a terminal.

Why Agencies Should Care:

This is what a modern, AI-native CMS looks like. You can sell this to clients who hate WordPress. Tell them: “Update your site from your notes app, and our AI will proofread and deploy it automatically.” It eliminates the devops bottleneck for content updates entirely, keeps the site blazing fast, and gives you a premium, low-maintenance infrastructure product to offer.

Built on NocoDB with n8n automation hooks. Tracks prospects, follow-up state, and engagement history without the overhead of Salesforce or the chaos of a spreadsheet. Fits in one screen.

Runs on a Hetzner VPS with Caddy, PostgreSQL, n8n, Open WebUI, and MinIO. Skills are markdown documents with JSON specs. The workbench is one endpoint that dispatches to any workflow by name. Saner than giving an agent root access. More powerful than most managed platforms.

When your web scraper feeds your prospecting tool, which triggers your CRM, which schedules your writing pipeline, which publishes to your site — that’s not a stack. That’s a flywheel.

This project documents and demonstrates the architecture of compounding automation.

Agents can search vault content, create notes, update frontmatter, and sync across sandboxes via Caddy and the ob CLI. The leenathan.com site itself is built on this — blog posts flow from Obsidian to Eleventy to production without touching a file manager.

Chains Exa semantic search and Firecrawl scraping through Claude to produce formatted PDF or DOCX reports on any company. Used for client research, competitive analysis, and prospect qualification. The kind of thing a research analyst charges for. Automated.

LidoDeck is the first concrete expression of the Purple Goo vision. Elixir’s concurrency model makes it a natural fit for agent swarms — lightweight processes, fault tolerance, and supervision trees that keep things running when individual agents fail. This is the long game.

Built on n8n and Claude, this system handles research aggregation, outline generation, draft writing, editing passes, SEO metadata, formatting, and delivery — all from a single trigger. The kind of thing a content team of five would own. One person runs it.

Targets EDGAR filings, navigates inconsistent formats across filing types, and outputs clean structured data ready for analysis or visualization. Built for real research workflows, not toy datasets.

Orchestrates Obsidian, n8n, Claude, and NocoDB into a unified system for tracking projects, priorities, and commitments. Morning briefs, weekly reviews, and routing logic handled by agents. The goal is high-signal attention on what matters, low-friction capture of everything else.