Business process automation, shown on real projectsMost of the stuff you repeat every week can now run on its own.
Reminders, invoices, lead lists, QR codes, prices from a marketplace, readings from a sensor, the weekly report. AI coding tools have made custom software cheap enough to build for a single process, so far more is worth automating than a few years ago. These are processes I have automated end to end, and two of them grew into products I run. Each one is small, logs every run and says when it fails.
| When | Task | Status | What happened |
|---|---|---|---|
| Mon 08:00 | Monthly invoice | done | Sent to the client, copy filed |
| Mon 09:15 | New website enquiries | done | Added to the CRM, owner told |
| Tue 07:00 | Competitor prices | check | One site did not load, flagged |
| Wed 18:00 | Weekly sales report | done | Emailed to you |
| Thu 07:30 | Supplier stock levels | done | Low-stock alert sent |
Products
Two products I run
Automations that grew into products. I build, host and support both myself.
- Remindlo: automatic SMS and email reminders for service businesses. Brings customers back for recurring services and cuts no-shows for garages, clinics, salons and vets in the UK. How it is built
- Scanlo: trackable QR codes. Free trackable QR codes that small businesses use to collect Google reviews and see which print actually gets scanned. How it is built
Projects
Other things I have automated
All projects-
Physical world + software
Heating, water and the house on live sensors
Software and hardware together: a cloud API, an ESP32 I built for the boiler room. It reads the heat meter, and a private chart site that shows what the heating actually did along with water usage, shades, electricity and daily costs.
Python · ESP32 / ESPHome · Kamstrup KMP (optical) · Viessmann API · Cloudflare Tunnel, R2, Workers, Access · Rain Bird local API · Xiaomi cloud (unofficial) · Telegram · GitHub Actions
-
Finance / compliance
Monthly e-invoices through a government API
A scheduled script that issues the same invoice every month through Poland's mandatory e-invoicing system, encrypts it, files the receipt and emails the PDF.
Google Apps Script · KSeF API 2.0 · node-forge (RSA-OAEP, AES-256) · Google Drive · Gmail
-
Sales / lead data
Contact emails from a list of company websites
A list of companies goes in, a CSV with ranked contact addresses comes out, with a status on every row so a failed site is visible rather than silently missing.
Python · requests + BeautifulSoup · CSV in / CSV out · parallel workers
-
Market data
Exporting every listing of a marketplace seller
Classic scraping found nothing on seller pages. The fix was to use the same data source the page uses, which also returned more fields.
Python · public JSON API · CSV / XLSX / JSON export
-
Content / AI in a controlled pipeline
AI-written news drafts, published by a human
Twice a week the pipeline collects facts, lets a model write from those facts only, checks the result and leaves a draft in WordPress. A person presses publish.
Node.js · OpenRouter (LLM) · WordPress REST API · Telegram · systemd timers
-
Sales / lead data at scale
Technology detection API for lead enrichment
Which tool does each prospect's website run? AI scrapers gave a different answer on every run. This loads the real page, watches what it requests, and returns the vendor with the URL that proves it.
Node.js · Playwright (Chromium) · Google Cloud Run (3 services) · Cloud Build · Clay
Approach
Every one of them has four parts
An automation nobody watches is a liability with a delay. Each project here is built from the same four parts, sized to the process.
The automation
Code that does the job: a scheduled script, a small API, a pipeline between tools a business already uses. Lightweight on purpose, with few moving parts and no per-seat licences.
Monitoring
Every run leaves a record. Failures send an alert to email, Slack or Telegram, and "it did not run at all" is checked too, because that is the failure nobody notices.
Reporting
A scheduled summary in plain language: what ran, what it produced, what needed attention. Weekly or monthly, wherever the owner already reads things.
Analysis
The data an automation collects is worth asking questions of. Ad-hoc answers when a question comes up, and a recurring analysis once it keeps coming back.
How it is built
Small, well-understood parts
- Cloudflare
- Workers, R2, Tunnel and Access: hosting at the edge, private dashboards without a VPN, servers with no open ports.
- Google Cloud
- Cloud Run for containers that scale to zero; Apps Script inside the Google Workspace a business already pays for.
- Supabase
- Postgres with access rules, edge functions and scheduled jobs, for anything that needs history.
- Python and Node.js
- Plain, tested code, versioned in Git and deployed on every push.
- ESP32 and ESPHome
- For when the data lives in the physical world.
Where AI fits
I write, test and document with AI coding tools. That is why a small automation takes days rather than months, and why automating a single process is now worth it.
Inside the automations, AI is used only where it is the right tool and always behind a check. In the news drafts pipeline, code collects the facts, the model writes from them, code validates the result, and a person publishes.
About
Who builds them
I am Peter Bednarek. I have been building on the web since 2004 and spent fifteen years running operations at LiveChat: support, HR, legal, administration and sales. Most of the processes I automate now are ones I used to do by hand.
FAQ
Questions
Can I use Remindlo or Scanlo without a custom project?
Yes. They are separate products with their own sites. You sign up there, and no custom work is involved.
Is the code public?
Some of it. Each project page says whether its repository is public or private. Private ones can be walked through on request.
Do I need a technical team or my own hosting for a custom project?
No. An automation can run on your own Cloudflare, Google Cloud or Supabase account, with access you grant and can remove, or I host and watch it and you only see the results and the reports. You can move from one to the other later, and in both cases you own the code and the data.
What happens when a website or API an automation depends on changes?
It fails loudly rather than quietly: an alert goes out, and the run log shows where it stopped.
Where are you based?
In Poland, on Central European Time: one hour ahead of the UK, with mornings that overlap the US East Coast. I work in English and Polish.
Have a process like these?
Alongside the products I take on a small number of custom automations. Describe what happens today and how often, and I will reply with questions, or with how I would approach it.
It can run on your own Cloudflare, Google Cloud or Supabase account, or be hosted and watched by me, so you only see the results and the reports. Either way you own the code and the data, and the price is fixed after a written scope.
Not a fit: a job an off-the-shelf tool already does well, scraping against a website's terms, or sending unsolicited messages in bulk.