Connect "I was able to try it" to the decision to introduce it.
OverviewExternal AI services cannot be used as is because they handle confidential information. Although it was verified, it has not been linked to on-site operations or existing systems. Link T&B confirms the target business and data handling, and supports implementation, evaluation, and operation, including the option of local LLM processing in your own environment.
Not only in model selection, but also in terms of operations, data, and evaluation methods.
- I want to use AI in work that handles confidential data.
- I want to connect the verification results to the introduction decision.
- I want to incorporate searching and summarizing internal documents into my work.
Before choosing a model, Decide on the work to be used
ApproachThe required data and evaluation methods differ depending on the situation in which AI is used, such as transcribing and summarizing meetings, searching internal documents, and assisting in answering inquiries. First, organize who will use it in what situations, and what you want to improve about your current work.
We separate the work that is left to AI and the work that humans have to check and make decisions. In addition to the naturalness of the text, we evaluate the accuracy and basis of the content. Decide how to check for errors and the conditions for refraining from using it.
Support Where You Need It
Support ScopeOrganizing operations, data, and terms of use
Check the information to be entered, user, storage location, and viewing privileges. In addition to confidentiality, we also organize on-site procedures and relationships with existing systems, and determine target operations.
Model/configuration selection and verification
Check candidate models and execution environments using evaluation data that matches the target business. Compare the conditions necessary to decide on implementation, such as response quality, processing time, simultaneous use, and operational burden.
Environment construction and collaboration with existing systems
Design processing locations, data passing, authentication, logging, and backups. When combining searches for internal documents, etc., also check the relationship between the documents that can be referenced and the permissions.
Evaluation and operation after starting use
Organize response confirmation, data updates, user guidance, and model update evaluations. We will consider a system that allows us to check the quality at the time of verification afterward.
Evidence to Decide Where AI Can Help
DeliverablesThe verification results will be compiled into materials for determining whether or not to introduce the system and the scope of its use. Below are examples of artifacts. The scope of delivery and confirmation method will be agreed upon at the time of contract.
-
Decide on verification conditions
Target operations, evaluation data, quality/processing time criteria
Example DeliverablesBusiness organization, verification plan, evaluation items -
Design processing and connectivity
Model, execution environment, reference documents, permissions, storage location
Example DeliverablesOrganizing the configuration plan, data handling and connection scope -
Decide whether to start using it
Verification results, scope of human verification, update method
Example DeliverablesEvaluation results, scope of implementation, usage/operation procedures
Start with a Pilot and Define the Rollout
ProcessWe assess whether summaries omit important information, answers cite the right documents, and the system responds appropriately when it cannot answer.
1. Determine work and evaluation conditions
Prepare typical inputs and expected outputs, and decide on criteria that suit your application, such as the basis for answers, errors, and processing time.
2. Try and evaluate candidate configurations
We verify the candidate model and execution environment, and organize the tasks that require human confirmation and the scope that can be left to AI.
3. Agree on implementation scope and operation
Based on the evaluation results, we will determine users, connection destinations, and update/maintenance methods. We will propose equipment and period according to the agreed range.
Map Where Data Is Processed, from Input to Storage
ConditionsEven when using local LLM, check the configuration of input acquisition, transcription, document search, log storage, management screen, update work, etc. individually. Design and specify which processing and data will reside within your environment and which will require external services and connections.
In addition to the location of data, we also check user authority, devices, data export, and operational management. Even when processing in your own environment, it is necessary to have a procedure for checking incorrect answers and to design permissions and operations to prevent information from being shared incorrectly.
5 steps to check data
This is an example of the flow of considering data handling. The scope that can be completed within the company and the presence or absence of external connections will be determined for each actual configuration.
We also check transcription, document search, log storage, and communication during updates, and clarify the scope of internal processing and the scope of external connections.
Tell Us Where You Want to Use AI
Getting StartedPlease let us know the work you would like to try, the type of information you handle, users, current equipment, and internal usage rules. Check to see if you can consider anonymized examples and data structures. When making a proposal, we will clearly state the scope of sharing of confidential information and the division of labor between our company and the partner.
Please let us know what business you would like to change with AI.
Contact Link T&B
Please tell us about the work you would like to use AI for and the conditions for the information you handle. You can consult with us about organizing the target business and a verification plan for small trials.
Discuss your AI projectIT/AI business materials
Download
You can view introductory materials for the IT & AI Solutions.
Request information →