Tokyo AI Solutions
Skip to main content

Secure RAG Solution

Keep confidential data in,
and surface internal knowledge with AI.

TAS Secure RAG is a RAG + generative AI development service that safely delivers a ChatGPT-like experience within your own closed cloud environment.

Reality Check

Common misconceptions about RAG adoption

Feed internal materials into ChatGPT and it gets smart right away

The moment you upload confidential data to a general-purpose SaaS, the information is sent to an external server. There are NDA-violation and terms-change risks, and you cannot step into the tasks where you truly want value.

Add AI and the effort of search disappears

Adopting only a tool without designing the mechanism means answers with an unclear basis cannot be trusted by the front line. Only when you design it to reference your own drawings and regulations with cited sources does it become usable in real work.

RAG is only for special large enterprises

In recent years, combinations of build templates and cloud have made it possible for mid-sized and small companies to adopt at a realistic cost. There is also room to use subsidies.

Readiness

If this sounds familiar, it is time to consult

The more confidential the task, the more important architectural design becomes — before tool selection.

Open for consult

You are reluctant to entrust confidential data and drawings to external services

A configuration that leverages AI while keeping data inside is possible. We design to fit your security requirements.

Open for consult

Veteran knowledge is concentrated in specific people

We convert tacit knowledge — drawing intent, response know-how, customer characteristics — into a searchable form. We aim for an internal AI dictionary that even new employees can reference.

Open for consult

You do not know where materials are, and time vanishes just searching

We integrate scattered information — specs, regulations, minutes — into a single knowledge base, retrievable instantly with cited sources.

Open for consult

You tried ChatGPT etc., but did not reach full use due to the confidential-data wall

There is an option to resolve the dilemma of “convenient but information control collapses” structurally.

Approach

What we value

In RAG adoption, the two most important things are building the structure that keeps information from leaking out first, and securing answer accuracy that the front line can trust. If either is missing, it becomes an unused system.

We take an approach of first confirming safety and answer accuracy in a small scope, then expanding gradually while reflecting the front line’s voice. This is a shared philosophy with our DX promotion support.

Design premise

Do not bolt security on afterward

Handling confidential data, security is a design premise. We build in the necessary configuration from the initial stage — encryption via KMS, traffic control via WAF, automatic data deletion after a fixed time, and more.

Verify small

Start with a specific department or task, and expand the scope only after confirming that impact occurred.

Make the basis visible

A state where not only the answer but also the referenced materials and basis can be confirmed, making it a mechanism the front line can easily decide with.

TAS GenAI Core

Our technology foundation

At the core of this service, we use TAS GenAI Core, an AI platform we have developed.

Based on the OSS “Gennai” published by Japan’s Digital Agency, it is designed to make minimal modifications to a quality-assured RAG platform and run stably. At the deal stage, you can experience a Gennai-based RAG environment and TAS’s proprietary security mechanism in a sales-demo environment that uses no real APIs or customer data.

Answers with cited grounds

Equipped with a RAG pipeline that references your proprietary data and answers while showing sources. Because the basis of an answer is verifiable, it can be shaped into a form easy to use on the front line.

Automatic answer-quality evaluation

Equipped with a reflection loop that automatically evaluates answer quality and regenerates answers that fall below standard. Aims to suppress what is known as hallucination.

Closed, encrypted, auto-deleted

Combines WAF, CMEK, S3, DynamoDB, and TTL to build in traffic control, encryption, audit trails, and automatic deletion from the initial stage.

The deployment cloud can be selected from AWS, Azure, or GCP to fit your requirements. Even when long-running processing is required, an asynchronous architecture that separates intake and processing runs stably without timeouts.

※ TAS GenAI Core consists of Gennai-derived code (MIT license) and our proprietary code (proprietary).

Process

Adoption flow

We deliver clear deliverables at each phase, keeping it always visible what changed.

Phase 1

Current-state grasp & security-requirement confirmation

1–2 weeksDeliverable: Issue list, target-data organization, security-requirement definition

We organize which tasks and which data to let the AI reference. Data confidentiality and constraints from NDAs and internal regulations are always confirmed at this stage.

Phase 2

Environment setup

Up to 1 monthDeliverable: Dedicated closed cloud environment, security configuration document

We cleanly build a private cloud area dedicated to you. The cloud used is selected to fit requirements; if a subsidy is used, we advance the application in line with this phase.

Phase 3

Knowledge learning (RAG build)

1–3 monthsDeliverable: Internal AI knowledge, answer-accuracy report

We ingest specs, drawings, contracts, etc., into a state that can answer with cited grounds. We confirm accuracy with real data before moving to full operation.

Phase 4

Rooting & company-wide rollout

From 3 monthsDeliverable: Operation manual, training, impact-measurement report

Starting from the department where impact occurred, we gradually expand to other departments — sales, development, general affairs. With screens and procedures that prevent users from getting lost, we aim for a state everyone uses daily.

Cost Support

About subsidy use

Secure RAG build and consulting costs may be eligible for digitalization / AI-adoption subsidies. Depending on the scheme, a substantial portion of the build cost may be covered.

Alongside Secure RAG development support, we can consult on subsidy eligibility and organizing information needed for application preparation. However, subsidies involve screening and selection is not guaranteed. They are also paid in arrears, so the adoption cost must be advanced.

About subsidy application support

Our subsidy application support is currently in preparation (business registration and support structure). We are accepting consultations on confirming schemes, assessing eligibility, and preparing for applications. Formal application representation and post-selection procedures will begin once our structure is ready.

See details on subsidy advisory →

Frequently asked questions

Q. Can we feed this data of ours into the RAG?

A. Suitability depends on the data’s form and confidentiality classification. In the initial hearing we hear the target data and tell you learnability and expected accuracy.

Q. Does the information really not go out?

A. We adopt a configuration in which processing completes within a closed cloud environment dedicated to you. We build in encryption, traffic control, and automatic deletion, designing so that a path to the outside is structurally not created. Specific requirements are confirmed individually.

Q. We have no IT lead — is that okay?

A. No problem. We accompany you through build, operation, and rooting. The operation screen is designed simply, and procedures are prepared.

Q. Can we start with just one task?

A. Yes. We recommend an approach that starts small with a specific department or task and expands after confirming impact.

Start with a free preliminary hearing

Can this data be used for RAG, and is it eligible for a subsidy? Tailored to your situation, we propose an individual estimate and configuration. The first hearing is about 30 minutes and can be held online.

Contact Us

If this resonates with you, please feel free to reach out.

Go to contact form