2015 · E.V.A.
A locally operated voice assistant developed to control a Windows computer without external funding: an early personal experiment in natural-language processing directly on the user’s machine.
Research
Experimental projects
Audit
Delivery
A continuous journey: from AI research to project delivery
Stéphane Gorius, founder of AIzyNow: more than ten years of artificial intelligence research and development, from E.V.A. to AIZYBRAIN, from scientific work published and archived at CERN through Zenodo to AI audits, project consolidation and system development.
AIzyNow is based in Alsace, France · engagements across France and internationally, remotely or on site depending on the project.
Why AIzyNow exists
I am Stéphane Gorius, developer and researcher from Alsace. For over a decade I have built systems where digital intelligence remains useful, traceable and, when needed, under your control, rather than reduced to an opaque cloud layer.
AIzyNow is not a generic label: it is the frame where I bring together R&D, experimental programmes and bespoke missions, with the same engineering bar, from audit to delivery.
The guiding thread: E.V.A. as early as 2015
Long before the rise of large generative models, I was already seeking to design useful digital assistance without depriving users of control over their data or computing environment.
In 2015, I developed E.V.A. (Extra Virtual Assistant), a voice assistant operating entirely locally and able to control a Windows computer, without external funding. It already represented an early personal experiment in natural-language processing directly on the user’s machine.
At a time when conversational assistants with modern natural-language processing capabilities were not yet widely accessible, E.V.A. was already raising several questions that continue to guide my work: memory, interaction, functional autonomy, local processing and control over data.
These early experiments gradually led to more complex architectures and eventually to the AIZYBRAIN research programme.
A continuity since 2015
From E.V.A. to AI project expertise
Each step fed the next: programmes, publications, prototypes, then a critical eye at the service of projects.
AIZYBRAIN
Research programme: hybrid cognitive architectures, persistent memory, functional continuity, identity and observability.
Publications
Reports and studies published and archived at CERN through Zenodo, with persistent DOIs, so the work stays verifiable.
Programmes
Prototypes and experimental applications, including AIZYBRAIN NANO, AIZYVISION and AIZYPLAY, testing principles in practice.
Audit & delivery
A methodological, technical and strategic eye at the service of AI projects: audits, consolidation and delivery.
Main programme
AIZYBRAIN: architecture, memory and continuity
The AIZYBRAIN programme is about hybrid cognitive architectures: persistent memory, functional continuity, functional identity, observability and behavioural evaluation.
I carefully separate what belongs to functional architecture, to behavioural observations, to hypotheses, to results and to interpretations.
Discover AIZYBRAIN« A functional architecture, uncertain behavioural observations; nothing is presented as proof of phenomenology. »
Functional architecture
Persistent memory, continuity across sessions, functional identity and observability designed as documented technical components.
Behavioural observations
Test campaigns document observed behaviours, their variations and their limits, without unwarranted extrapolation.
Hypotheses, results, limits
Everything is presented as an experiment or provisional interpretation. Phenomenology remains declared as indeterminate at this stage.
R&D & publications
The AIZYBRAIN work (reports, publications, research lines) is documented on the dedicated page. The scientific studies devoted to AIZYBRAIN are published and archived at CERN through Zenodo, with persistent DOIs.
From principles to applications
What I build today
Experimental programmes that confront AIZYBRAIN principles with practice.
AIZYBRAIN NANO
Functional experimental prototype: a local transposition of certain AIZYBRAIN principles, distinguishing local operation from external providers.
AIZYVISION
Experimental research programme: an assistive visual system designed primarily for people with visual impairments. Its purpose is to describe scenes locally, identify useful elements and support user autonomy, with particular attention to responsiveness and confidentiality.
Currently in developmentAIZYPLAY
Experimental multimedia application: transcription, translation and chat AI with on-device processing, with the same discipline.
Research & reports
The AIzyNow research centre gathers reports, publications and research lines around the AIZYBRAIN programme.
Research & knowledge sharing
Documenting artificial intelligence and making it accessible
Research has value only when it can be documented, discussed and shared.
Alongside my scientific reports, I published the French-language book L’Intelligence Artificielle : de ses origines à votre chatbot-IA personnel in 2023. It was designed as an accessible bridge between the history of artificial intelligence, its fundamental concepts and their practical implementation.
My publications, experimental programmes and auditing work pursue the same objective: making artificial intelligence systems easier to understand, verify and control.
From the builder's eye to the expert's eye
From research to audit: a critical eye
My work has also taught me to question: examining hypotheses, protocols, metrics and conclusions. This is what gives value to methodological and technical audits.
Audits are performed without conflicts of interest with the project under review: the audit states, describes and recommends; consolidation phases are then scoped separately.
See the audits on the homepageMethodological & scientific audit
Examining the coherence between hypotheses, protocols, data, metrics, results and conclusions.
Technical & architectural audit
Analysis of models, architectures, data flows, dependencies, performance and deployment context.
AI due diligence
Assessment of feasibility, maturity, risk and credibility before funding, partnership or deployment.
Governance preparation
Documentation, traceability, data and human oversight needed for governance and compliance preparation.
Beyond the assessment
Consolidate and deliver
From audit to delivery, the same engineering bar applies: clarify the need, strengthen the project, design the architecture and reach concrete deliverables: proof of concept, prototype or demonstrator.
A mission can remain standalone (an audit only) or serve as a first step of a wider engagement, scoped separately.
1
Assess
Feasibility, method, architecture, results, risks and claims.
2
Correct and consolidate
Methods, protocols, architecture, governance, documentation, reproducibility.
3
Deliver
Proof of concept, prototype, demonstrator, MVP and evaluation tooling.
4
Prepare
Publication, partnership, funding, fundraising or deployment.
Trust and protection
Your work, data and ideas stay yours
Missions can involve unpublished projects, proprietary architectures, research results, sensitive data or strategic information.
Confidentiality and the protection of intellectual property are part of every engagement's scoping.
Principles applied
- Non-disclosure agreement possible before sharing documents
- No publication, transmission or reuse without explicit authorisation
- No integration of client content into a training corpus
- No sending to a third-party AI service without prior agreement
- On-device processing preferred when technically possible
Regional roots, international engagements
From Alsace to France, Europe and beyond
The Alsace anchor makes regional expertise visible; it does not rule out national or international engagements. Audits and support can be delivered in French or English, remotely or on site.
Companies
Investors
Public sector
International
A method based on distinction and traceability
In both my research and auditing work, I systematically distinguish observed facts, experimental results, hypotheses and interpretations.
Protocols, evaluation conditions, limitations and areas of uncertainty are stated explicitly so that conclusions remain proportionate to the available evidence.
- Operation, protocols and results: documented and verifiable
- Hypotheses and interpretations: identified as such
- Limitations and areas of uncertainty: stated explicitly
Email contact
Let's discuss your AI project
Research, audit, consolidation or delivery: describe your context and I reply personally.
A first 30-minute scoping exchange, without commitment, can be arranged after this first contact by email.
Before sending sensitive documents: do not send any source code, confidential data or information falling under trade secrecy before the confidentiality framework is defined. A non-disclosure agreement can be signed beforehand.
Contact by email only.
Missions in France and internationally, remotely or on site depending on the nature of the project.