Calem Blogs

Blogs of CalemEAM

Calem AI - Part 6: Private AI for Your Maintenance Data

A technician can ask questions like "How was this issue resolved before?" in a work order and get an answer drawn from your own past work orders and equipment documents (screenshot below). The Ask AI feature in Calem uses a language model (LLM). There are potential issues with the use of public AI services. 

  • Maintenance records, asset details or site information are sent to ChatGPT, Gemini or Claude.
  • For some clients, internal policy of data security, privacy and compliance forbids it. 
  • For others, a data-residency rule in their country does.

1. Private AI

Private AI is offered by Calem as a solution so that your maintenance data is never shared outside of your private network (see diagram below). The language model (LLM) is hosted in your private network.

 2. Security and Privacy

The key advantages of private AI is its data security and privacy.

 3. Cost Analysis

Hosting an LLM requires premium hardware. A small model with 9B parameters with multi-language support works for Calem Ask AI. The model can be hosted on a CPU compute. For instance, one option is Akamai Cloud (Linode) Dedicated CPU, 8 vCPU / 16 GB, AMD EPYC 9845 supporting AVX-512 instructions. 

  • The cost starts from ~$200 a month for CPU compute. The response time is ~15 seconds for a query. 
  • The cost starts from ~$500 a month for GPU compute. The response time is ~1 second for a query.
  • The key advantage is the security and privacy of maintenance records and other data.
  • The cost is predictable regardless how many queries were asked.
  • Calem Ask AI works with generic LLMs since it is grounded on augmented-retrieval generation (RAG).
  • Organizations with own LLMs may re-use them with Calem to save cost. 
New Feature: Permit Approval by Department

Related Posts

By accepting you will be accessing a service provided by a third-party external to https://calemeam.com/