Client Report Generator
- Meta Ads + Google Ads + GA4 behind one adapter contract
- Config-sheet driven — new client = one spreadsheet row
- Draft-review mode: Gmail draft instead of auto-send
Human-in-the-loop approvals where money or reputation is at stake, deterministic core logic, and failure handling designed in rather than bolted on. The repo runs on n8n with custom code where a native node isn't enough — Make, Zapier or another stack when a client's setup genuinely calls for it, including the jobs that really are one simple zap. Every workflow below is a real, documented system — import it, press Test, and watch it route every branch with zero credentials.
Each ships with a problem statement, architecture diagram, setup path, input contract, and a demo lane exercised by CI on every commit. Not toy demos — the same structure delivered on paid work.
Weekly per-client marketing reports, end to end: three data sources behind one adapter contract, deterministic deltas, a human review seam, an idempotent send ledger.
Different businesses, same disease — repetitive work silently eating hours. Each profile maps to a working system in the repo.
| Marking | Meaning | |
|---|---|---|
| gate / approval | Human-in-the-loop approval gates where money or reputation is at stake | |
| retryOnFail · maxTries 3 | Retry/backoff on every credentialed node — transient failures absorbed | |
| appendOrUpdate · dedupe | Idempotent writes — re-runs are safe by construction | |
| quarantine lane | Suspicious records isolated instead of dropped or passed through | |
| ai at the edge | Deterministic core, LLM at the edges — it drafts, it never decides | |
| demo lane | Zero-credential test path — import, press Test, watch every branch |
Map your process, find the automation boundary.
You approve a written spec before anything is built.
Workflow + error handling + docs, shared workspace.
Tested on sample data, then a real pilot run.
You own everything — workflows, docs, credentials.
| Item | Part | Specification | Qty |
|---|---|---|---|
| BOM-00 | Automation audit | Send the task eating your week — get a written teardown naming what to automate first, with rough effort on each. See S-06. | free |
| BOM-01 | Workflow setup | Design, build, and deploy your workflow — n8n, Make, Zapier, or custom code; cloud or self-hosted. | per workflow |
| BOM-02 | n8n self-hosting | Docker/VPS deployment, queue mode, backups, monitoring. | per instance |
| BOM-03 | Workflow audit & rescue | Fix brittle or broken workflows; add error handling and retries. | per audit |
| BOM-04 | Document processing | OCR + LLM extraction on invoices, POs, intake forms — structured into your systems, low-confidence routed to a human. | per pipeline |
| BOM-05 | Custom integrations | Connect APIs that have no native nodes — Python/JS custom nodes. | per connector |
| BOM-06 | AI agent pipelines | LLM-powered classification, drafting, enrichment inside your stack. AI drafts — it never decides alone. | per pipeline |
| BOM-07 | Ongoing maintenance | Monitoring, alerting, versioned updates. | per month |
AVAILABLE FOR SHORT-TERM PROJECTS — WORKFLOW FIXES · NEW BUILDS · ERROR-HANDLING AUDITS
I'm Vinh Le — I design and build automation systems for small businesses, agencies, and solo founders. Send me 2–3 sentences about the process draining your time and the tools you run on. You get back a written teardown naming what I'd automate first, in order, with rough effort on each — free, by message, no call to sit through. If the honest answer is that a task shouldn't be automated, the audit says so.