Skip to content

Mike Matthews / Selected work

Software that connects
the work between systems.

I build custom applications, API integrations and AI workflows. These examples show how I turn an incoming event into a useful next step, with the checks and handoff needed to run it.

Built for my own business

Actual implementations, public workflow code and documented tests. These are work samples from Matthews Automation.

01 / Voice AI + software

A conversation becomes
an actionable request.

Twilio · LLM APIs · Next.js · Supabase

The problem
A phone call needs an answer, a useful record and a clear next step for the person following up.
What I built
My voice-agent work connects inbound call webhooks, conversational replies, call records and structured post-call summaries. I built the webhook routes, transcript handling, transfer routing and recording callbacks.
What you can inspect
Hank is the live AI receptionist for Matthews Automation. The public clip is a real example call performed by me, edited for length, with an excerpt of an actual delivered test alert. It is a demonstration, not a customer testimonial.
28-second edited demonstration. Illustrative stock footage; real Hank audio and a reformatted test alert.

Booking and other integrations depend on the connected systems and approved business rules. Their scope is confirmed before a build.

02 / Workflow engineering

From a missed call to a follow-up.

n8n · Twilio · Webhooks

The problem
A new inquiry arrives while the owner is busy. Follow-up needs to happen without losing the conversation or texting after a reply.
What I built
I built the intake workflow, contact state, timed follow-ups, inbound reply routing, quiet-hour handling and STOP/START/HELP handling. Owner alerts connect the workflow to the next human action.
Evidence and test limits
The public repository includes workflow exports, setup instructions and handset test notes. The notes separate verified paths from the two-phone bridge and other cases that still need verification.
Actual n8n lead pipeline with contact lookup, owner alerts, follow-up waits and reply checks.
Intake → owner alert → timed follow-up → stop on reply Actual workflow canvas. Open to inspect.

03 / Workflow engineering

Know when the intake system breaks.

n8n · Twilio API · HTTP monitoring

The problem
A broken website, low phone balance or changed routing can quietly interrupt new inquiries. Repeated alerts can create their own noise.
What I built
I built scheduled website and phone checks, remembered state, deduplicated alerts, quiet hours, retry queues and recovery notices. The workflow can run without a separate database.
Evidence and test limits
Ten live runs are documented, including website failures, recovery and failed-message retry behavior. Routing changes and carrier blocking were tested with mocks. The overnight morning brief was not verified live.
Actual n8n watchdog canvas showing website checks, Twilio account checks and notification decisions.
Check → compare saved state → alert once → report recovery Actual workflow canvas. Open to inspect.

Your next build

Start with the outcome.
Then prove one useful slice.

These examples are a starting point. I also take on custom portals, dashboards, data workflows and integrations where a ready-made tool does not fit the job.

  1. Define the result. Identify the systems, available access and what a successful run looks like.
  2. Agree on a paid first milestone. Test the difficult assumption before expanding the build.
  3. Deliver and hand over. Include error handling, relevant tests and clear operating notes.
Discuss a project with Mike ↗

Based in Marietta, Georgia. Available for remote projects.

Synthetic administration prototype

Source-linked drafts, conflicts and review changes.

Inspect a fictional structured-record example and 34 recorded fixture tests. No PDF/OCR, LLM/RAG, mining deployment or maintenance authorization is established.

Inspect the synthetic demonstration and its limits