One job, done well
The tool does one thing, so it is easy to judge whether it works.
Instead of an “AI for everything” I build small tools for one specific job: reading documents, sorting messages, writing summaries. A human approves the result, and when data is sensitive the tool can run on a local model.
The tool does one thing, so it is easy to judge whether it works.
Results show where the information came from, so you can check it quickly.
Important actions, such as sending a reply, need your approval.
When data cannot leave the company, the tool can run on a local model without sending content to the cloud.
We find one repetitive task where AI really saves time.
I build a working prototype and try it on your examples.
Together we judge result quality and decide where human review is needed.
I deploy the tool and document how to use it.
Analysis of Polish public tender documents with page-level citations: requirements, risks and a PDF report. Access by key.
View projectA character generator for novelists powered by a local AI model (Gemma 4). The model runs on my own PC, so content does not go to an external cloud.
View projectA bot that transcribes voice locally (faster-whisper) on your own GPU. No cloud, no audio leaves your machine.
View projectYes, which is why I design tools so results can be checked: citations to sources and human approval wherever a mistake is costly.
That is your choice. If data is sensitive, the tool can run on a locally hosted model. We agree this at the start.
We start with a process review for €35, and a simple automation with an AI step starts at €140 (starter prices). I quote custom tools after a conversation.
That is not the idea. The tools take over tedious, repetitive tasks; decisions stay with people.
The first conversation is free. Describe your situation in a few sentences and I will reply with a proposal and a quote.
Get in touch