Polcom AI as a Service
The AIaaS service provided as part of the Polcom AI Cloud facilitates the deployment of open language models using a specially optimised and dedicated IT infrastructure. This means that AI models can be deployed and scaled without the need for management of the hardware, system and operational layers.
AIaaS for organisations requiring control over data, models and costs
This service has been developed for organisations that wish to take advantage of the capabilities of generative AI, but want greater control over their data, models, runtime environment and costs. AIaaS can be used for a variety of purposes, including the development of AI assistants, chatbots, document analysis tools, process automation, content classification and the integration of language models with business applications.
Access to the service is provided in accordance with the OpenAI API standard. This facilitates integration with existing applications, tools and processes, whilst allowing for greater independence from public services offered by global providers.
Data privacy and sovereignty in AIaaS
In generative AI projects, it is particularly important to consider what happens to prompts, model responses, company documents and the data used in the inference process. Polcom AI Cloud has been designed to enable organisations to develop their own AI capabilities without unnecessarily increasing the risk of confidential business information being leaked.
The infrastructure located in Poland ensures data isolation and enables computing processes to be carried out in an environment separate from the public models of external providers.
Key principles
- No processing of prompts outside the service – query content is not analysed or stored for any purpose other than the execution of the computing session.
- No training of models on customer data – Polcom does not use customer data to train its own or public language models.
- Resource isolation – operations can take place within a sovereign Polish cloud infrastructure.
- Dedicated endpoints – organisations can reduce their reliance on public models and environments where the location and method of data processing are more difficult to verify.
This approach supports companies wishing to implement AI in areas requiring caution, such as the analysis of internal documents, the handling of personal data, and work involving technical, financial, medical, legal or strategic documentation.
Customisation and parameterisation of AI models
AIaaS allows the model’s runtime environment to be tailored to a specific use case. This allows organisations to better control performance, costs, context length, model type and query handling.
Configuration options:
- Model configuration – selection of architecture, model size, quantisation type and context window length.
- Performance optimisation – setting parameters such as KV Cache size, dtype and additional flags specific to selected models.
- Bring Your Own Model – the ability to run any open-source model supported by vLLM, including models that have been fine-tuned, e.g. using the LoRA method.