← All expertise

LOCAL AI · INFRASTRUCTURE, DOCUMENTS & USE CASES

Your knowledge.
AI on your infrastructure.

I bring together systems administration, application development and training to design an assistant suited to your environment. The project covers the models, authorised documents, access and onboarding for your teams.

YOUR NEED, PRACTICAL SUPPORT

What I can
take on.

INFRASTRUCTURE

Install the models

Select and configure models running on your resources, then verify their use through an interface suited to the project.

ACCESS

Define permissions and sources

Organise users, roles and document scopes. Authentication methods and any connection to your directory need to be defined.

DOCUMENT SEARCH · RAG

Your documents as a reference

The assistant searches authorised documents to enrich its answers. This is the principle behind RAG: connecting AI to your internal knowledge.

APPLICATION

Adapt the business workflow

Define the instructions, screens and documents needed for a business activity. Select features based on representative questions and tasks.

VERIFICATION

Check the answers

Test common questions, missing documents and access boundaries. Review the retrieved sources and errors before expanding use.

TRAINING

Support users

Explain the assistant’s capabilities and limitations, how to consult sources and the rules of use. Document update procedures.

FROM THE FIRST CONVERSATION TO ONBOARDING

Support
step by step.

  1. 01

    Define the scope

    Select a use case, its documents and the pilot’s acceptance criteria.

  2. 02

    Install

    Configure the models, document search, access and interface.

  3. 03

    Verify

    Test questions, review sources and check access boundaries.

  4. 04

    Share knowledge

    Document the installation, train users and plan ongoing support.

UNDERSTANDING RAG

Find information.
Keep a critical perspective.

A document assistant combines a language model with a search across a collection of documents: this is RAG. It selects passages relevant to the question, then uses those passages to formulate an answer. Choosing sources and respecting access permissions are part of the project.

For example, we might consider searching for an internal procedure or helping users consult technical documentation. These are potential uses to assess against your documents, not guaranteed results. AI can make mistakes: important answers must be checked against their sources.

BEFORE CHOOSING TOOLS

Start with your needs.
Define the scope.

Hardware, connectors and features are selected after reviewing your environment. These points form the basis of the scope and quotation.

DOCUMENTS

Which sources should we use?

Inventory files and their locations, and check their quality and currency. Scanned PDFs may require text recognition; tables and complex documents need testing. Supported formats are agreed after this review.

HARDWARE AND MODELS

Size the system before installation

Document volume, the number of simultaneous users and expected response times guide the choice of model, memory and any GPU. No standard server suits every SME: existing resources must be examined.

ACCESS AND CONFIDENTIALITY

Define the permitted uses

Define who can consult which sources, how users authenticate and which records to retain. Local hosting alone does not ensure security or compliance: access, backups and retention rules still need to be defined.

FROM PILOT TO EVERYDAY USE

Validate using your questions.
Prepare for what comes next.

An initial document scope allows the assistant to be tested against representative questions: does it find the right passages, respect access permissions and indicate when information is missing? Acceptance criteria are defined with you before any wider rollout.

The quotation must specify the expected elements: installation and configuration, connected sources, documentation, onboarding and acceptance conditions. It must also distinguish initial delivery from subsequent updates, backups, development and support work.

PREPARING OUR CONVERSATION

The questions
that matter.

Can the data stay within the organisation?

That is the aim of a local deployment. The scope must include the model, document search and associated components. Any external connections, updates or cloud features must be identified and discussed before they are enabled.

Do I need to replace my hardware?

Not necessarily. First, we need to examine the available resources and intended uses. A trial within a limited scope helps determine whether an upgrade or a dedicated machine is needed.

What budget and timeline should I expect?

They depend on hardware, document preparation, access and integrations. The first conversation helps clarify these needs to establish a scope and quotation; no universal price or timeline is stated here.

What information should I prepare?

Your priority use cases, number of users, document types and volumes, equipment and confidentiality constraints. To start, a description is enough: do not send sensitive documents through the contact form.

THE STARTING POINT

Let’s start with
your situation.

Tell me about your environment, the difficulties you face and what you want to improve. We can then define an initial practical direction.

Describe my needs