Plain automation is often enough. For reports, reconciliations and imports, rules, schedules and interfaces are all you need. AI comes in when a step has to read or generate texts, receipts or images. I tell you beforehand which case you have.
AI automation in Vienna. Built for continuous operation.
I develop AI automation in Vienna as custom software. Reports, reconciliations, imports and documents then run without manual work. Where a language model or an image model does a step better, I build it in, and a person signs off on the result before it is processed further.
I am a developer and consultant and I run what I build. If it is still open whether AI holds up in your case, we begin with a feasibility check: AI consulting in Vienna.
WHAT I AUTOMATE
I automated these business processes for clients, and I have maintained some of them for years.
Reports and reconciliations
Reports by email on a schedule, the reconciliation of delivery status and records, error emails on deviations. Built with Django and Celery.
Imports from other systems
Fetch new cases through SOAP or an interface, store attachments, create or update the user and the case, confirm receipt. A dry run shows beforehand what would happen.
Generating documents
Data turns into finished PDFs without anyone typesetting them. In one case study a generator replaces 45 minutes of InDesign per document with a run of three seconds.
Notifications without the flood
Every saved search runs on its own cycle, hourly, daily, weekly or monthly. A table remembers what was already reported, and each new result goes out exactly once.
Processing events reliably
Webhooks that lose nothing under load: database conflicts are retried, extra data is fetched only after the record is saved. In one case study that is more than 50,000 events a day, and none has been lost.
Backups and monitoring
Backups of database and files, monitoring of load and disk space, malware scans and update checks. An email report tells you what changed.
WHERE AI BELONGS IN THE WORKFLOW
AI is a step in the workflow, not the workflow. A model reads or sorts. A person confirms before it counts.
In a product of my own I built the reading of receipts with a language model: against a fixed schema, a person confirms every suggestion, a daily limit caps the cost.
Through an MCP connection, a standard interface for AI, an AI can read and create content. The rights of the signed-in role apply, texts are filtered, deleting is not possible. I built this for a client's WordPress site and for a product of my own.
AI AGENTS FOR COMPANIES
I develop AI agents for companies on request: programs that complete a task in steps, use tools for it and wait for sign-off before decisions with consequences.
What I can show are agents of my own. One fetches inquiries from a source on weekdays, scores them against a profile and files reports and drafts; sending is done by hand.
We start narrow: one task, fixed rights, sign-off by a person.
What you can build yourself and what running one adds is in the article Should I build an AI agent myself or have it built?
What I built myself and use
postkasten (opens a new tab)
Posts for LinkedIn and Bluesky live as Markdown files in a Git repository and go out on a schedule. A log prevents double publishing. No AI involved.
django-upgrade-report (opens a new tab)
Reads the lockfile of a Django project, queries PyPI and returns the order and the smallest safe step per package. No AI involved.
Keyword Planner MCP (fork) (opens a new tab)
Fork of an MCP server for the Google Ads Keyword Planner. I set country and language as defaults, added country codes and limited idea lists.
CUSTOM SOFTWARE INSTEAD OF N8N, MAKE OR ZAPIER
When a no-code tool is enough
If a workflow runs in a few steps between well-known services and nobody has to maintain it for years, a no-code tool such as n8n, Make or Zapier is often enough. Then I tell you so, and you save my fee.
When custom software is better
Does the workflow reach into your data and applications? Does it have to handle errors cleanly and run for years? Then I build it as custom software, for example with Python, Django and Celery. It has tests and monitoring, and your team can read the code.
HOW IT RUNS
First call
Free. You describe the workflow that is manual work today. I ask about data, errors and volume. Often plain automation is enough, and I tell you so right then.
The smallest useful step
I build the part of the workflow first that pays off on its own and put it into operation. You see early whether the direction is right.
Operation
The workflow runs with monitoring and alarms, and an error ends up in a report. Further steps follow once the first one has proven itself in operation.