Custom code
Past the node palette
When there's no prebuilt node, I write a custom n8n node or a small supporting service — API auth, pagination, rate limits, retries — instead of telling you it can't be done.
I'm a senior software engineer who builds automation the right way — n8n and Make workflows with real error handling, plus the custom code, API integrations, and AI-agent logic that no-code tools can't reach on their own.
Each of these ships as a real n8n workflow you can import and inspect — architecture, error handling, and the engineering decisions that keep it running. Not screenshots. Runnable systems.
Ask any question in Slack or Telegram and get an answer grounded in your own docs, every claim cited to its source file. The index re-builds itself whenever a document changes.
Teams lose hours hunting through Drive/Notion for answers buried in PDFs and docs. Generic chatbots hallucinate and can't cite anything.
Vector search over embedded doc chunks → an AI agent constrained to answer only from retrieved context, with source-file citations. Voice questions transcribed via Whisper. Delete-then-insert re-indexing keeps the store consistent.
Cited, grounded answers over private docs; zero manual re-indexing; multi-turn memory per user. Swap Telegram for Slack or a web widget without touching the retrieval core.
A lead hits the form and comes out the other end enriched, scored against your ideal-customer profile by an LLM, deduped, written to your CRM, and — if it's hot — pinged straight to sales.
Inbound leads sit un-triaged; reps chase cold ones and miss hot ones; the CRM fills with duplicates and blank fields.
Idempotent dedup → enrichment via a self-hosted caching microservice (fans out to Apollo/Clearbit/Prospeo, batches, controls cost) → LLM ICP-scoring to structured JSON → tier-based routing → CRM upsert → hot-lead Slack alert. Failures logged and surfaced.
Clean, enriched, scored records with no duplicates; hot leads surfaced instantly with reasoning; enrichment costs kept low via caching. Built to be observed, not babysat.
Drop in a podcast, webinar, or blog post. Get a LinkedIn post, an X thread, a newsletter section, and short-form video captions — on-brand, run past an approval step, scheduled, and logged.
One great piece of content should become ten. In practice it becomes zero — repurposing is tedious and gets skipped.
Transcribe/extract → AI agent distills "content atoms" → parallel channel-specific generators each carrying brand-voice guidelines → brand-safety filter → human approval in Slack → schedule/publish → log every variant to Notion/Sheets.
A steady content pipeline from assets you already have, with a human still holding the brand-risk lever before anything goes live.
The market is full of template-stitchers who hit a wall the moment a project needs real code. I'm an engineer first — the hard 20% is exactly what I'm for.
When there's no prebuilt node, I write a custom n8n node or a small supporting service — API auth, pagination, rate limits, retries — instead of telling you it can't be done.
Idempotency, dedup, retries with backoff, and failure alerting are built in from the start — so the automation doesn't silently rot two weeks after handoff.
RAG with citations, structured LLM extraction, function-calling, and evaluation — LLM steps wired into real systems, not toy chatbots.
Outgrown Zapier or fighting Make's limits? I rebuild flows in n8n properly — self-hosted where privacy or cost demands it — with the logic documented.
Scope a specific automation, build it tested and documented, hand it over with a walkthrough your team can actually maintain.
Fix a fragile scenario that keeps breaking, or migrate Zapier/Make flows into a clean, self-hosted n8n setup.
Design and build LLM-agent workflows — support triage, RAG assistants, research agents, content pipelines — with the plumbing to run in production.
Need an integration that doesn't exist? A published custom node or a small microservice bridges the gap.
Run an automation agency and hit a build you can't ship? I'm the senior engineer you subcontract the hard ones to.
Monitoring, tweaks, and new flows on a retainer — so your automations keep pace as the business changes.
Tell me the trigger and the outcome you're after, and roughly the volume. I'll tell you straight whether it's a quick build or a real project — and how I'd approach it.
levelbrookteam@gmail.comRemote · async-friendly · worldwide · GitHub ↗