Solutions
Custom business software with AI built in
Erudi builds custom AI solutions for very small businesses, SMEs and mid-market companies, in production within 2 to 6 weeks. Our team of AI engineers and researchers designs the software, then runs it for you.
Examples
What can custom AI do for a business?
These are the needs we come across most often. Each example is software built for one line of work and its data.
Internal chatbot over your documents (RAG)
Your teams ask a question in plain language. The system finds the relevant passages in your documents (hybrid keyword and vector search, then reranking) and writes an answer that quotes the source excerpt.
Document generation and tender responses
Starting from your templates and from content you have already approved, the AI writes a first draft in Word, Excel or PDF: technical proposal, report, meeting minutes. Your teams review and sign off.
AI agents that take on business tasks
An agent follows a procedure we define with you: it reads a case file, queries your tools through their APIs and prepares the next action. Sensitive steps wait for a person to approve them.
Your need doesn't look like any of these examples?
These examples are not a catalogue. We build the software you need, with AI at the steps where it helps. Tell us what you want to do: we'll tell you whether AI can do it reliably, and how we would go about it.
Our approach
Why does a project move faster with Erudi?
We assemble every solution from engines we have already built.
What is an engine?
A technical building block that comes back from one project to the next. We keep improving our engines and keep them at the state of the art. When the same need or technique shows up on several projects, we turn it into a new one.
For each client, we assemble these engines, then build what is specific to their business.
What you gain
- Lessons already learnedEach new project benefits from what we learned on the previous ones.
- 2 to 6 weeksto put a solution into production.
- A lower costthan building everything from scratch.
Your project is run by a small team of AI engineers and researchers. Decisions are made quickly, without a long approval chain for every feature.
How a project runs
How does a custom AI project work?
Understand your business and your data
We start by looking at how your teams work, with which documents and which tools. From that we write a scope, which also says what the solution will not do.
Build from our engines
We assemble our reusable engines, then develop what is specific to your case.
In production in 2 to 6 weeks
Before your teams use it on their own files, we evaluate the quality of its answers on your real cases. The timeline depends on the scope.
Run, maintain, improve
We host and maintain the solution, and your teams know who to contact. When a better model comes out or your needs change, we update it.
Want to know what this would look like in your company?
Write to usMaking the right choice
General-purpose or specialized AI: which should you choose?
ChatGPT, Claude and Copilot are useful, and sometimes they are all you need. The choice depends on your data, on the tools the AI has to work in, and on how reliable the results must be.
A general-purpose tool is enough when…
- you write, rephrase or summarize text that you provide yourself;
- each person uses it for their own tasks, with no process shared across the team;
- the information you need is public or fits in a few documents;
- you review everything, so an occasional mistake has no consequence.
Specialized AI becomes useful when…
- answers must come from your internal documents, with the source excerpt;
- the volume is more than you can paste into a conversation;
- the AI has to read from or write to your tools (CRM, ERP, databases);
- a whole team repeats the same task, and the result has to follow your rules and your document templates;
- you need to know where your data is processed, by which model, under which contract.
Operations
What does running your solution include?
Once the solution is in production, we run it. You have no server to manage and no account to open with an AI model provider.
- Hosting. Your application and your files have their own resources, with a dedicated database, in France or elsewhere in Europe, on infrastructure we operate.
- Compliance. We draft a data processing agreement (DPA) for your project, and our solutions comply with the GDPR and the AI Act. Security, GDPR and the AI Act: our approach
- Security. Your data is encrypted at rest and in transit, and isolated from other clients' data.
- AI model usage. It is included: you have no account to manage and no usage-based billing to track.
- Maintenance and updates. Fixes and updates, a switch to a better model when one becomes available, new features when your needs change.
- Support. Your users know who to contact when they have a question or a problem.
Security and compliance
How do we protect your data?
Hosting in France or elsewhere in Europe, an isolated environment for each client, encryption at rest and in transit, and no model training on your data. Every project has its own data processing agreement (DPA), and our solutions comply with the AI Act.
Our clients
Which companies do we work with?
Clients
We build AI solutions for very small businesses, SMEs and mid-market companies in Canada, Europe and North Africa.
Erudi operates from France and Quebec.
Frequently asked questions
What is the best AI for a business?
The best AI for a business is one specialized in its line of work and its data: it answers from the company's documents, applies its rules and works inside its tools. No single model is the best at every task. A model can write well and still be less reliable at extracting the figures from a table. Where the data is processed, confidentiality and the cost per request matter too. That is the kind of AI Erudi builds: we choose the models project by project, and the architecture lets us switch when a better one comes out.
Should a business use ChatGPT, Copilot, Claude or custom AI?
Microsoft Copilot works inside Microsoft 365, from the emails, files and meetings each user has access to. ChatGPT (OpenAI) and Claude (Anthropic) are general-purpose assistants for writing and analysis, and can connect to some common applications. None of the three follows your business procedures, or acts in software specific to your company, without configuration or development work. Custom AI is worth it when it has to work from your documents, act in your tools and follow your rules, or when you need control over where and how your data is processed. Both approaches can coexist in the same company. Compare general-purpose and specialized AI
How do you implement AI in a business?
Start from a frequent task whose result you can measure, rather than from the choice of a tool: finding information in internal documentation, for example, or producing the first draft of a report. Then connect the AI to the data of that task, evaluate its answers on real cases before opening it to your teams, and maintain it as models and needs change. With Erudi, a solution is in production in 2 to 6 weeks depending on the scope, because we start from engines we have already built. The timeline also depends on access to your data and on how long validation takes on your side. How a project runs
How much does AI cost for a business?
It depends on the form you choose. A general-purpose assistant is paid by subscription, per user; direct access to the models is billed by the volume processed; a custom solution combines an initial build fee and a monthly operating fee. At Erudi, both amounts are quoted per project. The monthly fee covers hosting, AI model usage and maintenance. What changes the quote: the number and type of data sources to connect, the volume of documents, the integrations with your business tools, how much human validation the process involves, and the hosting option (ours, your cloud or your servers).
How much does an AI agent cost?
The price of an AI agent depends mostly on how many tools it has to read or update, how long the procedure it follows is, and which steps need a human to approve them. There is the build (connecting to the APIs of your tools, business rules, tests on real cases), then the operations, since every task it handles uses calls to the models. At Erudi, an agent is part of a custom solution and is quoted the same way, per project. See the agent example
How do I know if AI can help my business?
AI has a good chance of helping if a repetitive task based on documents or data takes your teams several hours a week. Other signs: information nobody can find, large numbers of documents to read, teams already using ChatGPT with no rules in place. Describe the situation to us and we will tell you what looks feasible. Contact us
Do you need an in-house technical team to use AI?
No, not with a turnkey solution: Erudi designs, hosts and runs it. On your side, what you mostly need is people who know the business, to explain how you work and to validate what we build.
Does Erudi work on site?
Yes. Erudi operates from France and Quebec, and we travel to our clients when a topic is better handled in person, anywhere in Canada, Europe and North Africa.
Let's talk about what you need.
Tell us what you want to do. We'll tell you what is possible, and how.