AI workflows that assist people, not replace judgment

The useful automations are the boring ones: the quote assistant that explains a price, the OCR that attaches a signed delivery note to the right invoice, the lead that lands in CRM already assigned. I build those.

What you get

5 parts, one handover

  1. 01

    Claude API assistants that propose through tool calls while deterministic code does the arithmetic

  2. 02

    Lead staging from website, WhatsApp and email into Zoho CRM with round-robin assignment

  3. 03

    Document pipelines: delivery-note OCR and attachment, photo-to-PO drafts

  4. 04

    n8n and GitHub Actions workflows with monitoring and daily digests

  5. 05

    MCP servers so your team can talk to Books, CRM and Search Console from Claude

What it did last time

What it did last time

  1. A cost estimator covering 14 work types that pulls material cost from Zoho Books and produces a full quotation PDF

  2. A delivery-note system that attached its first batch of 59 signed notes to invoices and now runs daily

  3. A four-source lead staging app pushing reviewed leads into Zoho CRM

Questions

Questions I get asked

Which models and tools do you use?

Claude via the API and MCP for anything that needs language; n8n, GitHub Actions and Python for scheduling and data; Zoho Deluge where the logic belongs inside Zoho.

How do you keep AI from making costly mistakes?

The model proposes, code decides. In the estimating workbench the assistant explains and flags prices but never does the arithmetic, and every quote revision is immutable.

Contact

Let's build something fast.

Let's build something fast.

Roles, consulting, or a second opinion on a stack. One message is enough.