Local AI models for a production site
We deployed and trained two kinds of language model: chatbots that handle customer conversations and an internal assistant that takes routine work off the team. The models run on the company’s own infrastructure.
Problem
Part of every working day went on repetitive answers to customers and on searching for information inside the company. Handing correspondence and internal documents to outside services was not something the company was prepared to do.
Repetitive customer enquiries eat up time
Finding information inside the company takes too long
Data must not leave the company’s own network
Solution
We deployed two sets of local models and trained them on the company’s own material. Nothing goes out to external services: the models run on the company’s own hardware.
Customer-facing models: chatbots answer routine enquiries
Internal models: an assistant for the everyday work of staff
Models trained on the company’s documents and processes
Deployed on the company’s own infrastructure, with no external APIs
Stack and tools
Local LLMsSelf-hostedDockerTraining on company dataIntegration with internal systems
Results
2
sets of models: customer-facing and internal
100%
local: no data leaves the company’s own network
0
external APIs in the data path