AI outreach: an n8n, GPT-4 and Supabase pipeline
By Anurag Srivastav · AI Automation Engineer
I built an n8n and GPT-4 outreach pipeline that enriches lead data, drafts personalized emails, sends through SMTP and logs execution in Supabase. My portfolio reports 200+ automated emails per day. That is a reported sending-volume metric, not evidence of inbox delivery, replies, meetings or revenue.
Enrich context. Draft. Deliver. Track.Illustrative system map
The problem
Lead research, enrichment, message drafting and follow-up can become disconnected manual tasks. The project connected those stages into a workflow with a record of what had already happened to each lead, so processing did not depend on manually moving data between steps.
What I implemented
Built the autonomous n8n + GPT-4 lead-generation pipeline and automated email and WhatsApp follow-ups during my AI Automation Engineer role at Smartians AI. The documented stack includes Supabase, SMTP and Twilio.
Tools: n8n · GPT-4 · Supabase · SMTP · Twilio
Workflow at a glance
A simplified view of the stages described in the project account.
Enrich context. Draft. Deliver. Track.Illustrative system map
Collect lead-profile data
Explore the workflow
Choose a stage to follow the process.
Technical decisions and boundaries
Treat the model as one processing step
The existing n8n article describes a trigger, enrichment, model reasoning and an action stage. GPT-4 creates the personalized message while the workflow and connected services handle data movement and delivery. This division makes the intended responsibility of each component easier to understand.
Track progress per lead
My published implementation account describes a database state machine for each lead and logging in Supabase. The purpose is to make failed steps retryable without blindly restarting the entire process. The public material does not include failure-injection test results or proof that duplicate sends never occurred.
Separate throughput from business impact
A workflow can report a large send count without demonstrating successful delivery or useful conversations. Sending, delivery, replies and qualified meetings require separate measurements. No conversion or revenue lift is attributed to this project here.
Results and supporting evidence
A connected lead-enrichment and personalized outreach workflow; reported volume of 200+ automated emails per day.
What is documented
The sending-volume figure is reported in the original project description and n8n article. Raw delivery logs, the measurement period, unique-recipient counts, bounce rates and response rates are not published.
How to validate the outcome
To substantiate throughput, count unique message IDs accepted by the sending service in a stated time zone and date range, excluding retries and test messages. Report attempted, accepted, delivered, bounced and replied counts separately. These are validation definitions, not a claim that the unavailable historical logs have been audited.
Limitations
- The reported sending volume does not establish inbox placement, engagement or revenue.
- The public account does not contain production logs or a before/after labor-time study.
- Any reuse needs appropriate data access, recipient permissions, suppression handling and provider-limit checks; this page is not a ready-to-run campaign.
Project questions
What does 200+ emails per day mean here?
It is the sending volume reported in the portfolio. The measurement window and delivery logs are not published, so it should not be interpreted as verified delivery, replies or sales.
What was GPT-4 responsible for?
The documented workflow uses GPT-4 to generate personalized emails from enriched lead data. n8n coordinates the stages, SMTP sends messages and Supabase stores workflow state and logs.
How did the workflow approach retries?
The existing implementation article describes a per-lead database state machine so a failed step can be retried with context about prior progress. No duplicate-send benchmark or failure-test dataset is published.
Source material and related reading
These are my own published accounts, not independent third-party verification.