Glossary
Model fine-tuning
Fine-tuning means training an existing language model on a company's data so it picks up the company's style, terminology, and domain knowledge. The model itself changes, rather than just getting reference material with each question.
Why it mattersThe agent speaks your company's language and handles narrow topics better; a small fine-tuned model can run on your own hardware, offline.
Also called: Fine-tuning, Model fine-tuning, LLM fine-tuning, Fine-tuning a language model
How it works
How it works
You pick successful conversations from chat logs and call transcripts, strip out names, phone numbers, and other personal data, and show them to the model as examples. After several passes over the dataset, the model starts answering that way by itself. The result is always checked against expert reference answers – that shows whether the model actually got better.
- Dataset: successful conversations from your history
- Cleanup: no names, phone numbers, or other personal data
- Training: the model adjusts its internal parameters
- Evaluation: answers are checked against expert reference answers
Choosing
Fine-tuning or a knowledge base
Most of the time, a knowledge base is enough: the model finds the right article and builds its answer from it, and you fix mistakes by editing the article. Fine-tuning makes sense when a specific tone matters, there's a lot of niche terminology, or you need a small, fast model that runs without internet access.
In CONTACT.WORK
How it works in CONTACT.WORK
We train CONTACT.WORK agents in two ways: through the knowledge base, instructions, and reference answers without touching the model, or by fine-tuning it on your conversations. A fine-tuned model still uses the knowledge base, and we check every version against the reference answers. If your data can't leave your company's network, we fine-tune and run the model on your own hardware.
More
Where to use it
Backlinks
Where it appears on the site 14
Platform 2
Services 1
Glossary