Cor.株式会社 Cor.株式会社

Local LLM / secure AI PoC

We check whether generative AI can work even for tasks with information you can't send outside.

You have work that involves confidential data, so you can't send everything to a cloud AI. Yet you can't tell on your own whether a local LLM would really work for your business. Starting from there, this PoC gathers in three months the material you need to decide whether to go to production.

Common themes by industry

Every industry gets stuck for different reasons. Our industry pages cover the usual problems and how Cor. works alongside you.

  • Healthcare

    Organizing clinical data, patient response, information security, AI strategy

  • Professional services

    Contract review, applications & registration, deadline management, subsidy applications

  • Construction

    Labor management, design changes, safety checks, passing on know-how

  • Manufacturing

    Visual inspection, passing on know-how, equipment maintenance, inventory management

  • Education

    Test workload, parent communication, surveys & reports, ICT roster management

Is this where you are stuck?

  • You have confidential data, so you can't send it all to a cloud AI.
  • You can't judge in-house whether a local LLM would really work for your business.
  • With no internal rules, every trial ends at the PoC stage.

Terms used on this page

Local LLM
Generative AI that runs in your own environment, without sending data outside your company.
Secure AI
An AI environment built to handle information that is hard to give to an external AI, including the design of internal rules, permissions and logs.
PoC
Proof of concept: a small trial before production to check that something works.

How we proceed

Over three months, we proceed as follows. We don't bet on either cloud AI or a local LLM; the design assumes using each according to how sensitive the information is.

  1. Defining the target work

    We narrow down the work the PoC covers. If you have results from a confidential-data AI assessment, we set the scope based on them.

  2. Designing how data is handled

    We design where to draw the line on information given to AI (input boundaries), access rights, operation logs, and human approval.

  3. Building a test environment

    We build an environment for trying things out before production.

  4. Setting and measuring evaluation criteria

    We decide what counts as usable, then actually measure it.

  5. Deciding on production

    We decide whether to go to production.

What you have after the PoC

The goal is not a working demo but having the material in place to decide whether to go to production. Building something is not the goal in itself.

After the PoC

  • If it goes well, you move on to a production build and operations/maintenance (quoted individually based on requirements).
  • If not, we believe a decision of "not now" is also something you can take away as a result of the PoC.

Cor. keeps testing local LLMs and applies that experience to the design of each PoC.

Cost

Costs vary depending on the problem and scope. Our estimates make the assumptions, risks, and scope of work clear; the formal price and schedule are confirmed after requirements review.

Related work and case studies

See all work and case studies

Other services

Start by checking the information you handle.

Discuss a local LLM / secure AI PoC

The button takes you to our contact page, where our AI receptionist Cloudia will ask what you need.