AI consultant in Lille, France: from idea to AI in production
Feekr helps companies in Lille and across its metropolitan area (northern France) with generative AI: choosing the right use cases, building reliable agents and assistants, then running them on your cloud, with your teams.
An AI consultant based in the Lille metropolitan area
Feekr is based at 11B avenue de l’Harmonie, Villeneuve-d’Ascq, in the Lille metropolitan area. We work with SMEs and mid-sized companies in Lille, the Lille European Metropolis and the wider Hauts-de-France region (the northernmost region of France, bordering Belgium) that want to move from individual use of generative AI (an employee summarising a document or drafting an email) to reliable tools that are built into their processes, with known cost, quality and risks.
What sets us apart: we come from cloud infrastructure. An AI assistant that answers well in a demo is not enough. It has to respect access rights, be monitored, have its costs tracked and be maintainable by your teams. That is what we call AI in production.
What we build for companies in the metropolitan area
AI assistants on your documents (RAG)
Procedures, contracts, product sheets, technical documentation, reports: an AI assistant on your internal documents answers your teams’ questions, cites its sources and never shows anyone more than they are allowed to see. We measure answer quality against a set of real questions before go-live, and the assistant knows how to say it does not know.
AI agents connected to your tools
An AI agent goes further than an assistant: it acts within your tools (CRM, ERP, ticketing, email), for example to sort requests, prepare a reply or update a case file. We restrict its permissions to the strict minimum, require human approval for sensitive actions and log everything it does. Connections to your tools go through APIs or the MCP protocol, depending on your information system.
Putting AI into production and running it
Moving from a prototype to a service used every day takes LLMOps practices: automated evaluations on every change, prompt versioning, monitoring, alerts, cost per use and a fallback plan. This is often the step missing from AI projects that stall after the proof of concept.
The framework: usage policy, register and the EU AI Act
As soon as your staff use AI tools, you need clear rules: which tools are allowed, which data is off limits, human review, who is responsible for what. We help you put in place AI governance proportionate to your size and prepare for the obligations of the EU AI Act.
Our AI work
Our founder, Mohamed El Mardi, designs AI solutions for large customer experience groups. Without naming our clients, here is the kind of work we draw on:
- a document search assistant (RAG) for a business team, with embeddings, reranking, metadata filters, a confidence threshold and a refusal to answer when the source is missing, plus quality control of ingestion and a measured evaluation set;
- a FinOps assistant driven by an AI agent, which queries cloud costs through an MCP server and stops and starts non-production environments from a secure portal;
- a multi-tenant AI agent platform (Feekr R&D) in Python, with OIDC authentication, vector search on PostgreSQL and a streaming interface, covered by automated tests.
These projects are built on AWS, as Infrastructure as Code, to the same security standards as the rest of the infrastructure, including in environments subject to HDS (Hébergement de Données de Santé, the French certification for hosting health data).
Our method, from the first workshop to production
- Scope. A workshop, in your offices in Lille or by video call, to start from a specific business pain point, check the available data and security constraints, and choose a realistic first use case with measurable success criteria.
- Build. A first scope in real production, on your data, with pilot users. No isolated demo that never gets picked up.
- Industrialise. Infrastructure described as code, evaluations, monitoring, cost management, incident procedures and documentation.
- Improve. Measure usage and quality, adjust, extend to other use cases and transfer skills to your teams.
We stay neutral on models and providers: depending on your constraints, the solution relies on Amazon Bedrock, Azure OpenAI or conventional automation when AI adds nothing more.
Why a local AI consultant
AI projects are largely decided in conversations with business teams: understanding a task, watching how a team works, checking the assistant’s answers with the people who know the subject. Being a few kilometres away means these sessions can happen on site, in Lille, Roubaix, Tourcoing, Marcq-en-Barœul or elsewhere in the metropolitan area, and that we can come back quickly when a decision is stuck.
The metropolitan area also has a useful ecosystem for getting to grips with AI, such as EuraTechnologies or La Cité de l’IA (see our page on cloud and AI consulting in Lille). Our role begins when an idea has to become a solution in production. For companies elsewhere in the region, see our work across Hauts-de-France.
Book a call
A 15-minute call is enough to tell whether working together makes sense. Email us at hello@feekr.io, call us on +33 3 20 84 14 27 or book a call.
Frequently asked questions
What does an AI consultant do for a company in Lille?
They help you choose the AI use cases that are worth pursuing, check that the data you need exists and is accessible, design the solution (assistant, agent, automation), put it into production on your cloud with the right guardrails, then train your teams to run it. At Feekr, the same team covers scoping, build and infrastructure.
Are you an AI agency or a consulting firm?
Feekr is a technical consulting firm. We do not resell software or place contractors: we scope, build and industrialise AI solutions with your teams, on AWS or Azure, then make you self-sufficient.
Do you come to our offices in Lille?
Yes. We are based in Villeneuve-d’Ascq and come on site in Lille and across the metropolitan area for scoping workshops, demonstrations and training sessions. Build work is then mostly done remotely, with regular check-ins.
Does our data stay with us?
That is the option we favour for sensitive data: models are called from your own AWS or Azure account, in a European region, with your access rules and your logging. The exact choice depends on your constraints and is made during scoping.
How much does a first AI project cost?
It depends on the scope, the state of your data and how deeply the solution has to integrate with your tools. We do not quote generic prices: after a scoping workshop, you receive a written proposal with a scope, deliverables and an estimate of the running cost per use.
A cloud or AI project to scope?
15 minutes to understand your context and tell you honestly whether we can help. Based in Lille, France, working on site or remotely.