n8n workflow automation for your business
I build n8n workflows that connect the tools you already use — website forms, email, Telegram, WhatsApp and APIs — and keep a person in the loop where a decision matters. As a full-stack software engineer, I can also extend your own backend when a ready-made node is not enough.
Processes I automate
Each example links to a workflow I built. They show the approach; they are not claims of measured time savings or client results.
- Website Lead Capture & Telegram Alerts: validate form submissions, send each one to Telegram once, and record whether delivery succeeded. The contact form on this site uses it.
- Outreach Email Approval via Telegram: queue drafts, approve or skip each one from Telegram, and send only approved emails through SMTP.
- WhatsApp Audio Replies & Automated Follow-Ups: answer incoming WhatsApp messages with prerecorded audio in a set order, then send scheduled follow-ups.
- AI Support Email Triage & Reply Drafts: classify incoming Gmail messages, search a knowledge base, and prepare reply drafts that stay unsent until reviewed.
Tools these workflows connect
Every integration below appears in the workflows above. Other services can usually be reached through their APIs with n8n’s HTTP Request node; that is checked against the service’s documentation at the start of a project.
- Webhooks for website forms and incoming messages.
- Telegram Bot API for notifications and Approve / Skip buttons.
- Gmail for reading support email, creating drafts and adding labels.
- SMTP for sending approved email.
- Evolution API for WhatsApp messages and audio.
- An AI model with knowledge-base search for grounded reply drafts.
- n8n Data Tables for contact records, queues and delivery logs.
- JavaScript Code nodes for validation, fingerprinting and batch summaries.
Human approval and failure handling
A workflow that acts on your behalf needs clear rules for uncertain cases. These safeguards are already part of my workflows:
- Approval before sending: outreach emails wait for an Approve or Skip decision in Telegram, and only the owner’s decision counts.
- Drafts, not replies: AI-prepared support answers are saved as Gmail drafts with a review label.
- Manual handling when sources are missing: if the knowledge base has nothing relevant, the email is flagged for a person instead of drafted.
- Duplicate protection: message IDs, content fingerprints and submission claims stop the same item being processed twice.
- Send locks and suppression lists: a scheduled worker handles one batch at a time and skips suppressed addresses.
- Recorded outcomes: deliveries are logged and reported, and an uncertain send is not retried automatically.
How a project runs
- Map the process: what starts it, which tools are involved, who decides, and what should happen when a step fails.
- Agree the scope: triggers, approval points, the data stored, and the accounts each step needs.
- Build and test with sample or anonymized data before connecting live accounts.
- Hand over the workflow in your n8n instance, with notes on what each step does and how its credentials are set up.
Hosting, subscriptions and API usage
Workflows run in an n8n instance you control: n8n Cloud, or n8n self-hosted on your own server. Accounts for the services a workflow uses — such as Telegram, Gmail, a WhatsApp provider or an AI model — stay in your name, and their subscription and usage fees are paid to those providers, separately from my work.
Credentials live in n8n’s credential manager, not in workflow text. Monitoring, updates and changes after handover are agreed separately rather than assumed.
Questions
Do I need my own n8n instance?
Yes. The workflow runs in an n8n instance you control, either n8n Cloud or a self-hosted installation. The right choice depends on your budget, your data-handling requirements and who will maintain the server.
Can an automation keep a human approval step?
Yes. My outreach workflow sends each draft to Telegram for an Approve or Skip decision, and my support inbox workflow keeps every AI-prepared reply as a draft. These examples show the approach; they do not establish a measured time saving.
What happens when a step fails?
That is decided during planning, step by step. My workflows record each outcome, flag items for manual handling, and avoid automatic resends when it is unclear whether a message was delivered.
What should I share when requesting an automation?
Describe the current process, the tools involved, a sample of the data with private values removed, and who needs to approve what. Share credentials later through an agreed secure channel, never through the contact form.