Tool-based training
AI training on your internal tool
Your organization has built its own generative AI assistant: a web interface based on open-source models, a portal connected to the Claude or OpenAI APIs, and a multi-model tool hosted in-house. We train your teams directly on it, as we do at CNES, BMS, and Airbus, using your models, rules, and the tool's actual limitations.
Training on your internal AI tool, in brief
Nine lines to find out if it's right for you. Architecture, method, price and financing are detailed below.
- What it is
- A in-house training in generative AI animated on the assistant that your organization has developed or integrated, and not on a commercial tool.
- For who
- Large organizations, public agencies, manufacturers and laboratories that have deployed a homemade tool and want it to be truly adopted by their teams.
- Tools covered
- WebUI interfaces on open source models, portals connected to the API from Claude, OpenAI or Mistralassistants self-hosted multimodels.
- Duration
- De 1 hour (Prompt Party®) to a multi-wave programpreceded by a half-day of hands-on training on the tool by our trainers.
- Where and in what language
- At your sites in France or remotely for your teams abroad. French or English.
- Data
- Demonstrations on your tool, in your information system : no prompts or documents pass through a public account.
- Price
- €1,500 to €2,000 excluding VAT half-day sessions for up to twelve people €2,000 to €3,000 excluding VAT In-house day, route available upon request.
- Funding
- Eligible for OPCO and ESF+ funding. NextStart is a registered training organization (n ° 11756140675).
- Lead time
- First response and proposal within 24 business hours, without obligation; detailed proposal after a 30-minute scoping exchange.
Our benchmarks on internal tools
Three programs conducted on in-house assistants, each with a different architecture, documented in our customer cases.
Why a homemade tool requires separate training
An internal assistant addresses the confidentiality issue, but it creates an adoption problem. Employees compare it to the consumer version they use at home, don't know which model to choose, encounter a limitation they don't understand, and revert to their old habits, or worse, to a public tool. Generic training on ChatGPT or Copilot doesn't address any of these issues.
- Choose the right model depending on the task: a quick open source model for reformulating, a more powerful model for analyzing a long document, when the tool offers several.
- Know the real limitations of the tool: context size, accepted documents, lack of web navigation or connectors, quotas, conversation history.
- Write prompts that work on open source models, which are often more sensitive to the structure of the instruction than mainstream models.
- Apply the rules of your charter: what can go into the tool, what must remain out of any AI, how to check an answer before using it.
Three in-house tool architectures that we are familiar with
Every architecture has its strengths and weaknesses. Our support is tailored to the one your IT department has chosen.
A WebUI interface based on open source models
The CNES and CEA model: an open-source chat interface, several open templates hosted internally, and a choice left to the user. The training teaches how to choose the template, structure instructions, and share the best prompts within teams.
A portal about the APIs of Claude or OpenAI
The BMS model: an internal conversational assistant connected to the Claude and OpenAI APIs, alongside Copilot. The training explains what changes compared to the consumer versions, which tool to use for which task, and how to leverage the latest models within a controlled framework.
A self-hosted multi-model assistant
Airbus's model: a customized, multi-model tool built on open-source models and hosted within the group's network. The training works on recreated cases, compliant with confidentiality and export control requirements, with an internal expert co-facilitating the sessions.
Is your tool based on another model, Mistral for example, or on an agent platform like Dust or Prisme.AI? The approach is the same: we first get to grips with the tool, then we build exercises on what it can really do.
What we did on internal tools
Three references, three architectures, with their figures. The details of each program are in the client case study.
CNES · SidonieWebUI and open source models, since 2024
Discovery webinars have become a recurring program, from the Prompt Party withAgent PartyEach session begins with the agency's IAG charter and all demonstrations take place on Sidonie, in Toulouse, Paris and French Guiana.
BMS France · Claude API and OpenAIAn internal assistant and Copilot, in six waves
Eighty-four Prompt Party and fifteen Agent Party in person, with ten internal champions trained before each wave, to get the internal conversational assistant and Copilot adopted by the headquarters teams.
Airbus · Open source multimodelThe GenAI Challenge on the group's assistant
Six Prompt Party In English for France, Germany, Spain, and the United Kingdom, all demonstrations focus on the secure, proprietary, and internal assistant, co-presented by an Airbus expert. No public tools are used.
How we prepare training on your tool
We don't discover your tool on the day of the session. Three steps, carried out with your project team or your IT department, ensure that each exercise works in your real environment.
Test access, a list of available models, activated functions and known limitations: our trainers test each use case on your tool before teaching it.
The exercises are based on situations from your profession and what the tool can actually do, not on demonstrations of a consumer tool that wouldn't work for you.
As at Airbus and CNES, the tool expert answers technical questions while we lead the practical exercise, then your ambassadors take over. Free AI Ambassador Kit.
Three levels, from the first prompt to the delegation
The same formats as on the tools on the market, adapted to the functions that your tool actually offers.
Adopt
Prompt Party® (1 hour or half a day) and Masterclass : to make the internal tool a daily reflex, for each job case.
Create assistants
Agent Party® : if your tool allows the creation of specialized assistants, your teams leave with their own, connected to their reference documents.
Industrialize
Coding with AI et Agent Studio : for teams that use the tool's APIs and want to build business applications on top of them.
Details of formats, durations and deployments are on the page Our offers.
Security, sovereignty and compliance
In-house tools often originate from a security or sovereignty requirement. Training adheres to this requirement from beginning to end and makes it understandable to every employee.
- Everything happens within your information system: no demonstration on a public account, no customer documents transmitted to a third party.
- Usage charter: we open each session with yours, like at CNES with the IAG charter, and we make it a reflex rather than a constraint.
- Export control and intellectual property: exercises on fictitious or public documents, validated by your security teams.
- Article 4 of the AI Act: training materials and certificates that document your employees' AI proficiency.
Price benchmarks
The same prices as other tools on the market. The exact quote depends on the number of sessions, customization to your specific cases, and the time it takes to learn how to use the tool.
This training is eligible for OPCO and ESF+ funding and can be charged to your skills development plan, provided you submit the documents required by your funding body. A written proposal will be sent within 24 business hours, with no obligation.
FAQ
Can you provide training on a tool you've never used?
Yes, that's the case for most in-house tools. We request test access a few weeks before the session, a list of available templates and functions, and we test each exercise on your tool before teaching it. The tool's in-house expert can co-lead the demonstrations.
Our tool has fewer features than ChatGPT: is that a problem?
No, that's precisely the purpose of the training. We demonstrate what the tool does well, how to work around its limitations with better guidelines, and why it remains the right choice for your data. Employees stop comparing it to the consumer version and learn how to use it effectively.
Are open-source models efficient enough for professional use?
For most writing, summarizing, and document analysis tasks, yes, provided the instructions are structured and the right template is chosen. During the session, we demonstrate the concrete differences between templates and the cases where it's better to switch to a more powerful one, when your tool allows it.
Our data must not leave the information system: how do you manage this?
Everything takes place on your tool, in your environment. The exercises are based on your real situations but on reconstructed or public documents, validated with your mentor, and the rules of your charter are reviewed at the beginning of each session.
Our tool evolves every month: does the training remain valid?
Yes. We teach methods that outlast model changes, and we update the materials with each wave. On long programs, like at CNES, each session takes into account the version of the tool in use.
How much does training on an internal AI tool cost?
€1,500 to €2,000 excluding VAT for a half-day session for up to twelve people, €2,000 to €3,000 excluding VAT for a full-day in-house session. Multi-phase programs, including tool training, internal support, and adoption measurement, are available upon request, with pricing decreasing based on volume.
How to finance the training?
The training is eligible for OPCO and ESF+ funding. NextStart is a registered training organization (no. 11756140675): the session is part of your skills development plan, with the agreements and documents expected by your funding body.
What happens after my request?
Flavien Besset will personally respond within 24 business hours with an initial proposal. A 30-minute scoping meeting will then allow him to understand your tool, your teams, and your constraints, before the detailed proposal and training agreement are finalized.
Train your teams on your internal tool
Two fields and a sentence are all you need: click what applies to you, the sentence will write itself, complete it if you wish (tool name, available models, teams involved, deadlines).
Do you prefer to call or write? +33 1 89 16 48 05 or hello@nextstart.aiYour contact details are only used to reply to you.
Do your teams also use a commercial tool?
Many organizations combine an internal assistant with a market tool, such as BMS with Copilot. We provide training on both, with a consistent learning path.
NextStart is a training organization registered under number 11756140675. This registration does not constitute state approval.