AI for business: AI that solves real problems in your business

We don't start with the technology, but with the process that takes up most of your hours. These are the cases where AI tends to add value first in a business, and none of them is a generic chatbot.

What we've built

100% on-premise

Intelligence analysis engine for defense

An engine for the defense and military domain that turns large volumes of information into structured intelligence: which actors are involved, what each one does, who they're connected to and how a conflict evolves. It starts from open sources (OSINT) and, because it runs entirely inside the client's infrastructure, it's designed to also work with internal and classified information.

  • Defense and security
  • OSINT
  • 100% on-premise models
  • PMESII framework

Why on-premise

In defense, information is too sensitive to send to anyone's cloud.

  • No data passes through external services like OpenAI, Anthropic or Google
  • Can run on air-gapped networks, with no internet connection
  • The client controls the models, the data and the logs
  • Lets you cross-reference open sources with information that could never leave
  1. 01

    PMESII classification

    Each text is placed across the political, military, economic, social, information and infrastructure dimensions.

  2. 02

    Actors and relationships

    A specialized model extracts entities and relationships, and a language model refines them: it resolves aliases and references (“the aggressor” → the named actor) and finds the ones that were missing.

  3. 03

    Military module

    Identifies the parties to a conflict and classifies deployments, shows of force, threats and use of force.

  4. 04

    Escalation or de-escalation

    Measures whether each source pushes toward negotiation or toward escalation, and how it sizes up the threat.

  5. 05

    Disinformation and sources

    Detects propaganda narratives, geopolitical framing and whether claims are backed up or simply taken as fact.

  6. 06

    Measurable answers

    Every question to the model returns an option, a score or a probability, not free text: results can be aggregated, compared and audited.

We've also built

Back office

Digitizer for delivery notes, receipts and invoices

The problem
Documents arrive on paper, as PDFs or as phone photos, and someone has to type their data in by hand.
What we built
A system that reads them automatically and extracts the supplier, date, line items and amounts, ready to review and post into your business system.
  • Azure Document Intelligence
  • Data extraction
  • Integration

Finance

Assistant for a financial analysis platform

The problem
A platform with lots of financial information about each client, but getting value from it takes knowing where to look and how to read it.
What we built
An assistant built into the platform that answers with each client's own data: you ask in plain language and it replies with their own figures, explained, instead of generic answers.
  • Language models
  • Each client's data
  • Built into the platform

Other ways AI can help you

Customer service

Emails and tickets, sorted

Today
The shared inbox fills up and every message is assigned by hand.
With AI
Classifies by topic and urgency, routes it to the right person and suggests a draft reply for a person to review.
What we need from you
Access to the inbox or the ticketing system.

Operations

Demand forecasting

Today
Purchasing is decided with a spreadsheet and gut feeling: too much of some things, not enough of others.
With AI
A model learns from your history, seasonality and calendar, and retrains itself on new data.
What we need from you
Sales history, ideally two years or more.

Sales

Customers about to leave

Today
You find out a customer has left when they simply stop coming back.
With AI
Scores each customer's churn risk and alerts you in time to act.
What we need from you
Purchase or usage history and past cancellations.

Team

Meetings and calls nobody forgets

Today
What gets said to a customer stays in the memory of whoever took the call.
With AI
Transcribes, summarizes, pulls out the to-dos and saves it all to your CRM.
What we need from you
The recordings and the tool where you want everything stored.

Management

Alerts when something doesn't add up

Today
Errors in orders, payments or consumption are discovered at the end of the month.
With AI
Monitors your data continuously and alerts you as soon as something is out of the ordinary.
What we need from you
Access to the data you want to monitor.

Production

Quality control with computer vision

Today
Inspecting parts or products is manual and depends on the eye of whoever checks them.
With AI
A vision model detects defects from photos of the product.
What we need from you
Photos of good and defective products.

When a generic model isn't enough

We train and fine-tune custom models

Some problems can't be solved by plugging into an API. When your data is highly specific, when you need precision or when the information can't leave your company, we build a custom model and keep it running in production.

Predictive models

  • Time series for demand, sales and inventory
  • Customer classification and scoring
  • Anomaly detection
  • Recommendation systems

Training and fine-tuning

  • Fine-tuning language models on your data
  • LoRA / QLoRA adapters
  • Distillation and quantization to cut running costs
  • Evaluations designed for your specific case

In production

  • Automatic retraining on new data
  • Monitoring model quality over time
  • Usage costs measured and under control
  • Self-hosted models when the data can't leave

Sensitive information? We can also set all of this up on-premise, without your data ever leaving the company. See on-premise AI

Special requirements

On-premise AI: your data never leaves the company

If you work with confidential information, we set up AI inside your own infrastructure. We use open models that run on your servers, so your documents and data never pass through third-party servers like OpenAI, Anthropic or Google.

Built for sectors with sensitive information: defense, healthcare, legal, finance, industry or government. It's how our intelligence analysis engine for defense works.

What's included

  • Open models (Llama, Mistral, Qwen…) on your servers or in your private cloud
  • Works offline, if you need it to
  • Assistants and search over your documents, all on-premise
  • Access control: each person only sees what they're allowed to
  • Model updates when better versions come out

Good to know first

  • · You'll need a server with a GPU (your own or in a private cloud). We'll help you choose and size the hardware.
  • · Open models are somewhat less powerful than the big commercial ones; for most business uses they're more than enough, and we'll tell you what to expect in your case before we start.

FAQ

What people usually ask us

Is my data used to train other companies’ models?

If your data is sensitive, we set up AI on-premise: it runs on your servers and the information never leaves your company. If we use cloud services, we choose options that don't train on your data, and we explain it to you before we start.

What's the difference between on-premise AI and using ChatGPT?

With ChatGPT, your information travels to another company's servers. On-premise, everything runs inside your infrastructure, even without an internet connection. The trade-off: you need a server with a GPU, and open models are somewhat less powerful than the big commercial ones.

Do I need a lot of data to get started?

It depends on the case. An assistant that answers from your documents works with the ones you already have; a predictive model needs historical data, ideally a couple of years' worth. We'll tell you before we start.

What is fine-tuning, and when is it worth it?

It's adjusting a model with your own examples so it learns your vocabulary, your format or your judgment. It's worth it when a generic model can't reach the accuracy you need, or when you want a smaller, cheaper model that does one specific task very well.

Will AI replace my team?

We use it to take repetitive work off their plate and help them make decisions. For sensitive tasks, like replying to a customer, a person reviews everything before it goes out.

How do I know whether AI will work for me?

Start with an assessment: tell us about the process, we'll analyze it and tell you honestly whether AI makes sense there or whether there's a simpler solution.

Not sure which case to pick? Start with an assessment: tell us how you work and we'll advise you on where AI would save you the most hours.