Six services. One method underneath.

Operations automation build

You have a process held together by copy-paste, memory and good intentions: leads moved between tools by hand, follow-ups that depend on someone remembering, statuses that never sync. I map it end to end, then build the n8n or Make engine that runs it.

The builds are not demos. A duplicate trigger does not create a duplicate record, a mid-run failure resumes instead of restarting, and anything that needs judgment stops and waits for a person instead of guessing.

What you get

A running engine, documented, with alerting

Runs on

n8n or Make.com, your existing tools

Built in

Idempotency, audit log, human gates

Proof

Cold email engine, AutoRecruit

Data layer and migrations

Automation is only as reliable as the data underneath it. I design Airtable and Supabase schemas that automations can trust, write sync scripts that keep two systems agreeing, and run migrations that arrive verified rather than hoped-for.

Airtable to Supabase migration

For when you outgrow the base. The schema, the data and the automations that depend on them move together, and the result arrives verified.

Make to n8n migration

For teams that want their automations on an engine they control. The workflows move without changing what they do.

What you get

A schema, sync scripts, or a verified migration

Runs on

Airtable, Supabase, Postgres

Built in

Idempotent syncs, field-level ownership

Migrations

Airtable to Supabase, Make to n8n

Internal tools and dashboards

When operating data lives across five tools, every basic question becomes a research project. I build Next.js apps and KPI dashboards that give a team one place to operate, fed by scheduled syncs rather than manual updates, so the numbers are current and there is exactly one definition per metric.

What you get

An internal app or dashboard, live data

Runs on

Next.js, Vercel, your data layer

Built in

One definition per metric, no manual refresh

Proof

AutoClose, AutoRecruit portal

Systems integration

Your CRM, billing, outreach and data tools each work fine alone and disagree constantly together. I connect the stack with custom API integrations built to survive bad days: rate limits respected, retries with backoff, failures logged and alerted instead of swallowed.

What you get

Integrations that hold under real traffic

Worked with

HubSpot, Stripe, Slack, Instantly, Snov, Apify and more

Built in

Retries, rate-limit handling, failure alerts

Proof

Sponsor sourcing pipeline

AI in the workflow

I add language-model steps inside real workflows: classification, generation, routing, retrieval and agents. The discipline is what makes them safe to run unattended: answers grounded in your sources instead of invented, oversized inputs filtered before they cost money, anything the model is not confident about routed to a person.

What you get

AI steps that run unattended, safely

Runs on

OpenAI, Cohere, Supabase pgvector

Built in

Source grounding, cost ceilings, confidence gates

Proof

Recruitment retrieval layer

Automation audit and rescue

You already have automations. Some are breaking, some are switched off because nobody trusts them, and the person who built them is gone. I trace what a workflow actually does end to end, classify every node by rule into core, redundant, dead or unknown, and hand back a documented system with a defined path to production.

Rescue means you keep what works. The goal is a system someone other than the original author can operate, not a rebuild invoice.

What you get

A node-by-node map, a verdict, a repair path

Runs on

n8n, Make

Method

Every node classified: core, redundant, dead, unknown