Automation and AI

Work that removes a manual step, and the machine learning and vision work that has to run on hardware too small to be generous about it. The test for anything scheduled here is simple: does a failure tell you it failed, or does it leave a silent gap in the data.

01

What this looks like in practice

Automation work has one recurring shape. Somebody is doing a sequence of steps by hand, often across several tools, and the sequence has grown past the point where a person should be doing it. The job is not writing a clever script. It is reading the actual sequence, noticing which steps carry weight, and removing the ones that were only there because of an old limitation.

The failure mode is well known. A script gets written to do one thing, it works, nobody records the assumptions it makes about the systems underneath, and two years later it fails in the middle of the night for a reason nobody documented. A cron job with no error reporting is worse than a person doing the task by hand, because it also takes away the reminder.

02

Work done in this area

n8n and Slack integration
Troubleshot a Slack connection inside an n8n workflow that had stopped delivering. Small job, done the same day.
Stack
n8n, Slack, webhooks
Raspberry Pi vision classifier
An emotion classification system set up and running on Pi hardware with OpenCV, which meant choosing a model the hardware could run rather than the best one available.
Stack
Python, OpenCV, Raspberry Pi, machine learning
Voice clarity enhancement
Signal processing to improve intelligibility on degraded recordings, with the reports delivered alongside the work.
Stack
Audio processing, signal processing, machine learning
YouTube to Icecast pipeline
A scripted path pushing audio from source to an Icecast mount, built so it could be run repeatedly without supervision.
Stack
FFmpeg, Icecast, shell, Linux
Retrieval tooling
Download and collection tooling built to a stated scope and delivered as a quick, clean piece of work.
Stack
Python, Linux
Load balancing across services
Traffic distribution set up so a failing backend drops out rather than taking the service down with it.
Stack
Linux, Nginx, networking
Survey and reporting workflow
A recurring research and reporting process automated around an n8n instance, run on a schedule with results delivered automatically.
Stack
n8n, research workflow, reporting
Docker to GitHub deployment
A dynamic website deployed to GitHub with the surrounding configuration handled, delivered on time with some extra fixes done without being asked.
Stack
PHP, Docker, GitHub, cloud computing
03

Questions

Is it worth automating a task that happens once a week?

Sometimes, and the deciding factor is not frequency so much as whether the task has to be done correctly every time. A weekly task that takes twenty minutes and occasionally gets the formatting wrong can be worth doing properly. A weekly task that takes two minutes and has never gone wrong is not worth a project.

Will the script keep running on its own?

Yes. Anything scheduled gets error reporting, so a failure tells you it failed instead of leaving a gap in the data. That is the part most automation projects skip, and it is the part that decides whether you trust the result six months later.

Can you work with the tools I already have?

Yes, and that is usually where the useful work is. Making something you already own do what it was bought to do costs less than replacing it, and an n8n instance with one broken connection is a better starting point than a new platform.

Do you build AI features?

Where they earn their place. The work has included vision systems on constrained hardware, audio processing, and workflow integrations. The part I take most seriously is where the system has to be right rather than impressive, because a confident wrong answer costs more than no answer at all.