Local LLM / AI platform PoC — Putting the material for a decision in your hands

Challenge

There is work that involves confidential data, so sending everything to a cloud AI is out of the question. At the same time, whether a local LLM — one that runs on your own side without sending data outside the company — would really hold up in your business is not something you can judge alone. We hear this more and more often. A PoC also has a way of ending at the point where something has been built: you hear plenty of stories of a working prototype that never went on to production.

Approach

What Cor. Inc. aims for in a PoC is not a demo that runs, but a state in which the material for deciding whether to go to production is in place. We do not commit in advance to either cloud AI or a local LLM; we design on the premise that each will be used where it fits, according to the sensitivity of the information involved. We are consistent about not making the building itself the goal.

Implementation

Over three months, we work through the following.

• Organizing the target operations
• Designing how data is handled (input boundaries, access permissions, operation logs, human approval)
• Building a test environment
• Setting evaluation metrics and measuring against them
• Deciding whether to go to production

Where an assessment (the Confidential Data AI Assessment) has already organized this ground, we use it as the base and narrow the scope of the PoC accordingly.

Results

At the end of three months, the material for deciding whether to go to production is in your hands. If it goes well, you move on to production build-out and ongoing operation (quoted individually according to requirements). If not, we believe that deciding not to do it for now is also something worth taking away as a result of the PoC. Cor. Inc. carries out validation work on local LLMs continuously, and feeds what accumulates there back into how we design PoCs. The fee is from ¥3M over three months (varying with the problem and the scope). If this is on your mind, starting with the Confidential Data AI Assessment is one way in.