As a freelance ML engineer I design, train and measure models where an off-the-shelf API isn't enough: LLM fine-tuning, custom classifiers, embeddings for search and evaluation of AI systems. I work in Python and PyTorch for companies in Slovakia and Czechia.
I train a custom model only when an off-the-shelf model measurably falls short on your data.
Who it is for
- Companies that need to predict customer churn, demand or risk
- Teams whose off-the-shelf model isn't accurate enough on their data
- Companies that want to measure whether their AI actually works
How it works
- We set a metric and a simple baseline to compare against.
- I prepare the data and train the model, from simple to more complex.
- I evaluate it on data it didn't see during training.
- I deploy it behind an API and set up quality monitoring in production.
What you get
Baseline measurement
How well an off-the-shelf model or your current solution works on your data.
Model or fine-tuning
A classifier, embeddings or LoRA fine-tuning, depending on the task.
Evaluation
A comparison with the baseline on a test set you keep.
Handover
Code, training procedure and documentation so you can retrain the model.
In practice
In my public ai-ml-learning repository every step has a baseline, a metric and a results table: the churn model reached a PR-AUC of 0.664 against 0.265 for a naive model, and LoRA with a third of the parameters reached 95.9 % against 96.9 % for full fine-tuning.
- ClaimPilot
Insurance claims: 99.3 % of critical fields correct.
What changes in the company
The decision about a custom model rests on a number, not a hunch, and the model can be retrained without me.
Technology and data security
I process data as agreed, locally or in your cloud. Results are reproducible: environment, code and metrics.
Price
€35–45/h or a fixed price based on scope after a consultation. Pricing
FAQ
When is LLM fine-tuning worth it?
When you need a consistent style, format or narrow task that the model handles poorly even with a good prompt. For knowledge from company documents, RAG is the better fit.
What is LoRA?
LoRA is a fine-tuning method that trains only a small part of the parameters. It is cheaper and faster than full fine-tuning, and the result is usually very close.
Do I need a lot of data?
It depends on the task. First we check whether the task can be solved without training, and only then prepare data so we can measure whether the model helps.
How much does an AI project cost?
Private AI over your documents starts from €490, an AI audit costs €540–1,080, automating one process €800–2,400, and production RAG or an AI agent prototype €3,600–5,400. My hourly rate is €35–45 and the first 30-minute call is free.
Related
- Services
- Local LLM
AI on your own server. Data stays in the company.
- RAG for companies
AI that answers from your documents and cites the source.
- AI audit
Where AI pays off for you and what it will cost.