Skip to content
Discuss your project

Software architecture · Applied AI

The software you need. Built by AI agents, directed by an architect.

Business application, custom tool, AI agent or migration: IntelliNova builds with AnythingMachine, its AI agent factory. A senior .NET architect makes the decisions; every delivery is tested then reviewed, in the technology you want. More than twenty years of critical software: the kind that must keep running when everything else stops.

Discuss your project Explore our expertise

Orchestrate. Verify. Deliver.

Human supervision

Architecture · decisions · validation

  • Code agent

    Implementation

  • Test agent

    Acceptance tests

  • Review agent

    Independent review

hmmlearn-net / HMM-13

611
passing tests
Blocked
first review
Approved
after correction

Documented example: review of 7 October 2026.

Explore the actual work

Ars dirigit, machina exsequitur.

Human vision. The power of agents.

Experience grounded in practice

Software, from design to production.

  • 20+ years

    of critical software: public and European institutions, public alerting, telecoms in crisis situations. When everything stops, our tools keep running.

  • .NET expert

    our specialty. With AI and our coding experience, we also deliver in the technology you require.

  • Local AI

    models run and compared on our own infrastructure.

  • Since 2003

    projects in healthcare, transport, agriculture, telecoms and online payments.

01 / Expertise

Eight disciplines. One standard of care.

A new project, an existing application or a concrete use for AI: we connect your business goals with sound technical decisions.

  • .NET architecture and development

    Architecture, APIs, distributed systems and performance: a .NET expert who makes decisions and stays hands-on. Thanks to AI, we also deliver in the technology of your choice.

  • Custom business applications

    Software built around your processes, from data to interfaces, with incremental deliveries and clear acceptance criteria.

  • Applied AI and agents

    Integrate language models and automate tasks within your tools, with explicit boundaries and human approval.

  • Local AI models

    Select and evaluate a model on your tasks, then run it on your infrastructure when confidentiality requires it.

  • Custom AI for decision-making

    Got data? We build your own model, using deep reinforcement learning or another approach, to help you make the right decision.

  • Application takeover and modernisation

    Audit existing software, improve its reliability or migrate its technology while preserving useful business behaviour.

  • Web and integrations

    Accessible websites and web applications connected to your APIs, data and the services your business relies on.

  • Audit, quality and performance

    Understand what slows your application down, assess the risks and identify useful improvements before committing to the next step.

Explore the services

02 / In practice

An open project. Work you can inspect.

The hmmlearn-net port and its public archives show how our approach translates into code, tests and review decisions.

Scientific port · public repository

hmmlearn-net

Porting Markov models from Python to C# requires preserving numerical results. This internal project combines a Python reference, parity tests and documented reviews. Its code and history are available to inspect.

  • Python → C#
  • .NET
  • Numerical parity
  • Open source
29
implementation and hardening tickets
5
milestones documented in the forge

HMM-13 archive · 7 October 2026

When a review changes the outcome.

A real episode from the port: an issue identified after the tests had passed.

  1. VerificationPASS

    The tests pass.

    The first review records 611 passing tests. That result does not end the code review.

  2. Review agentREQUEST_CHANGES

    A calculation error is found.

    The reviewer identifies a mathematically incorrect fallback in Gaussian sampling and blocks integration.

  3. Pilot decisionDECISION

    The scope is clarified.

    The pilot orders the fallback removed. A failed factorisation now raises an explicit error. The code is revised.

  4. Second reviewAPPROVED

    The correction is approved.

    The reviewer verifies the removal, reruns the checks and all 611 tests, then approves the delivery.

Counts and decisions from the archived review dated 7 October 2026. Detailed sources are available in the public repository.

Our development environment

AnythingMachine

Our software factory organises specifications, agents, verification and reviews. The architect supervises the work and makes significant decisions. The hmmlearn-net project provides a documented example.

  • Orchestration
  • AI agents
  • Tests
  • Human supervision

Ars dirigit, machina exsequitur.

03 / Our approach

Accelerate the work. Keep control.

AI agents write the code; an architect decides and answers for the result. Every task follows the same five steps, from need to delivery.

  1. Define

    We agree with you what gets delivered and the conditions for accepting it.

  2. Test

    Tests are written before the code: they describe the expected result.

  3. Develop

    AI agents write the code in an isolated environment.

  4. Verify

    Automatic build and tests: nothing moves forward if they fail.

  5. Review

    A second, independent review, then a documented delivery.

Every delivery is traceable: specification, changes, tests and review. Architecture decisions remain human.

04 / Working together

A technical partner, from start to finish.

  • Decide together

    A reasoned diagnosis, clear trade-offs and an architecture that fits your needs.

  • Build incrementally

    Defined milestones, demonstrations and acceptance criteria to track progress.

  • Protect your data

    Tools, models and where they run are chosen according to your confidentiality requirements.

  • Prepare what comes next

    Maintainable code, useful documentation and a handover that supports your application’s future.

Let’s talk about your next project.

An application to build, a system to stabilise or an AI use case to explore? Let’s start with your context and the outcome you need.

Discuss your project