The Future API Stack for Mass Innovation

Welcome and Opening Remarks

Why Open Source Matters... and Even more now with AI

Jean-Baptiste Kempf

CTO - Scaleway

Why LLMs make me disagree with (almost) everyone

This presentation critically examines the current discourse surrounding Large Language Models (LLMs), challenging prevailing narratives from both enthusiasts and skeptics. Drawing from a dual perspective as an AI governance professional and a machine learning researcher specializing in generative AI, the talk bridges regulatory concerns and technical realities. It focuses on distinguishing between hype and actual potential of LLM technology, while examining the disparity between imagined risks and actual harms associated with LLMs.

The discussion offers evidence-based insights into the ethical and practical challenges in the field, introducing effective practices for LLM implementation with emphasis on agentic workflows. By providing a grounded analysis that challenges common assumptions, this talk aims to contribute to more effective strategies for navigating the complexities of LLM development and governance, encouraging a reevaluation of current approaches in both areas.

Djalel Benbouzid

Core contributor - AI, Algorithmic and Automation Incidents and Controversies (AIAAIC) organization

Eyes Wide Open, AIs Wide Open?

Yann Lechelle

Co-Founder and CEO - Probabl.ai

Moshi : A Speech-Text Foundation Model for Real-Time Dialogue

I will present moshi.chat, the first real-time conversational voice AI ever released.

Neil Zeghidour

Chief Modeling Officer and Co-Founder - Kyutai

AI and the Future of Humanity : Biopunk or Biohacking? A Panel Discussion

Alain Damasio

Sci-fi Author

Arnaud Auger

Director - Cathay Innovation

Building with GenAI for Public Service

Ulrich Tan

Head of DataLab and Deputy to the Director - ETALAB

AI Adoption : Engaging Minds, Developing Skills, Delivering Results

In this presentation, we will explore how Groupe Rocher is integrating AI through a comprehensive approach that focuses on acculturation, training, and supporting business teams, ensuring that every initiative aligns with the company’s overall strategy. We will discuss how we critically assess the use of AI, prioritize relevant use cases, and empower internal skills to achieve tangible and sustainable results.

Yannick Tremblais

AI/IT Innovation Manager - Groupe Rocher

GenAI : Fake It Until You Make It

How we leverage LLMs to re-invent the search and matching process at Malt. What benefits with GenAI, what workarounds with the new challenges.

Claire Lebarz

Chief Data & AI Officer - Malt

Gemini for Workspace - The future of work

Issam Idjerouidene

Google Workspace Specialist - Google Cloud

Putting the 'Engineering' into Prompt Engineering

As a prompt engineer people always tell me it isn't real engineering, or that it's going away in five years, or even that it's already dead. In practice the top model companies and frameworks invest heavily and promote prompt engineering techniques, and new papers are released every weeks showing gains from prompting. There is a big difference however between what Twitter gurus or the average ChatGPT user is doing, and what a prompt engineer actually does day to day to earn the engineering title. To call it engineering you have to be testing how LLMs work at scale, and I'll walk through several practical examples of how to do that, as covered in my book on prompt engineering for O'Reilly and my work on client projects and my own products.

Mike Taylor

Prompt Engineer - Author of "Prompt Engineering for Generative AI: Future-Proof Inputs for Reliable AI Outputs" O'Reilly 2024

Speech To Speech : An Effort for An Open-Sourced and Modular GPT4-o

This talk introduces Hugging Face's "Speech-to-Speech" project, an innovative open-source pipeline designed for natural speech interaction with LLMs. The project integrates advanced components such as Voice Activity Detection (VAD), Whisper models for Speech-to-Text (STT), Language Models (LM), and Parler-TTS for Text-to-Speech (TTS) synthesis. We'll explore the pipeline's architecture, emphasizing its modularity and potential applications in areas like real-time translation and accessibility. Attendees will gain insights into the technology's capabilities and future impact communication with LLMs.

Andrés Marafioti

Machine Learning Research Engineer - HuggingFace

API Roulette: Betting on the Right Provider in the GenAI Gold Rush

Jean-Louis Quéguiner

CEO - Gladia

Leverage the Agentic Stack in My Apps

Frédéric Wickert

Technical Specialist - Microsoft

"Trends of AI" Voice of Users

Exclusive reveal of the first cross-functional impact study on AI in business, conducted by the Les EnthousIAtes collective in partnership with KPMG France. Four key functions scrutinized by 80 "C-level" executives from HR, IT, Finance, and Marketing. Get ready for insights tailored to specific sectors but, more importantly, with a game-changing cross-functional perspective! Don’t miss this chance to be ahead of the curve!

Etienne Lecoeur

Co-Founder - Les EnthousIAstes

From MLOps to GenOps,A Panel Discussion with the GAIN (GenAI Network)

Anne-Laure Giret

Head of Google Cloud AI GTM, EMEA South - Google

Matthieu Boussard

Head of R&D - Craft AI

Paul Mochkovitch

Co-Founder & Tech Lead - Molia

Samya Barkaoui

GenAI Research Lead - Pigment AI

Etude French Tech IA

David Chriqui

CEO - Mozza

Building an AI-first culture in your organisation

Guillaume Simon

Staff Product Manager - Photoroom

David Chriqui

CEO - Mozza

Dounia Zouine

Co-Founder - MEUF (App)

How AI enables mapping the Anthropocene and informing public policies

In this keynote, IGN's Director General will demonstrate how artificial intelligence is revolutionizing environmental monitoring and territorial analysis in France. The presentation will detail IGN's strategic transformation into a leading AI-powered institution, with ten operational AI systems now supporting critical environmental policies. Through concrete examples like the Zero Net Artificialization (ZAN) monitoring and the FLAIR initiative, the talk will illustrate how AI enables unprecedented insights for public decision-making in ecological transition. The discussion will highlight IGN's journey to becoming a key AI player in the French public sector, supported by a team of 30 AI specialists and significant investments in training data. The presentation will conclude by exploring how these technological advances contribute to building France's digital twin, enabling data-driven environmental policies and climate adaptation strategies.

Sébastien Soriano

Directeur Général - IGN 🇫🇷

The Challenges of Creating a Competitive AI Infrastructure in Europe

Adrienne Jan

Chief Product Officer - Scaleway

How AI can serve the environment ?

Gilles Babinet

Author of "Green AI" and Co-Chairman - Conseil National du Numérique and Digital Champion pour la France à la Commission Européenne

How do we navigate the gap between oversight of information production, responsible governance, and data

Imane Idrissi

Technorealism - Adeo

Arnaud Billion

Phd in Computer Science, AI Researcher - IBM

Henri Albert

Co-Founder & CPO - ROTI.express

Is AI ready for patients? Lessons from 2000 medical conversations with "Mo"

Antoine Lizée

Data & Engineering - Alan

Thinking about thinking: Metacognitive Capabilities of LLMs

Reinforcement learning from human feedback (RLHF) is a go-to solution for aligning large language models (LLMs) with human preferences; it passes through learning a reward model that subsequently optimizes the LLM's policy. However, an inherent limitation of current reward models is their inability to fully represent the richness of human preferences and their dependency on the sampling distribution. In the first part we turn to an alternative pipeline for the fine-tuning of LLMs using pairwise human feedback. Our approach entails the initial learning of a preference model, which is conditioned on two inputs given a prompt, followed by the pursuit of a policy that consistently generates responses preferred over those generated by any competing policy, thus defining the Nash equilibrium of this preference model. We term this approach Nash learning from human feedback (NLHF) and give a new algorithmic solution, Nash-MD, founded on the principles of mirror descent. NLHF is compelling for preference learning and policy optimization with the potential of advancing the field of aligning LLMs with human preferences. In the second part of the talk we delve into a deeper theoretical understanding of fine-tuning approaches as RLHF with PPO and offline fine-tuning with DPO (direct preference optimization) based on the Bradley-Terry model and come up with a new class of LLM alignment algorithms with better both practical and theoretical properties. We finish with the newest work showing links between and building on top of them.

Michal Valko

Principal Engineer, LLama Team, GenAI - Meta

Gemma : Open Models Based on Gemini Research and Technology

In this talk we will discuss Gemma, a family of lightweight, open-access language models (2B-27B parameters) built upon the technology behind Google's Gemini. Gemma achieves state-of-the-art performance in language understanding, reasoning, and safety, offering competitive alternatives to models that are 2-3 times bigger

We detail Gemma's development and highlight its strong performance across diverse benchmarks. We also discuss our commitment to responsible release, providing comprehensive safety evaluations and pretrained/fine-tuned checkpoints. Gemma empowers the community with powerful and accessible LLMs, fostering innovation and safer AI.

Thomas Mesnard

Research Scientist on Gemma/Gemini - Google Deepmind

They Said It Couldn't Be Done : Training SOTA LLMs On Fully Open Data

Presenting world's first suite of models trained on exclusively open data with permissive licences

Anastasia Stasenko

Co-Founder and CEO - pleias

SmolLM: The Rise of Small Language Models

On-device large language models (LLMs) are revolutionizing AI by making advanced models accessible in resource-constrained environments. This talk covers SmolLM, a series of compact yet powerful LLMs, focusing on our training process, data curation, and ways to leverage these models for on-device applications.

Loubna Ben Allal

Research Engineer & lead SmolLM - HuggingFace

Model Extraction : Why You Should Care and How to Protect Yourself?

Are you deploying on-device AI algorithmes (IoT, mobile, desktop, on-prem)? They are vulnerable to reverse-engineering: a competitor can locate, extract and copy your models in a couple of minutes.
In this talk, we'll talk about model extraction, the consequences for your company, the techniques for reverse-engineering a model and the dedicated techniques for AI models protection.

Marie Paindavoine, PhD

CEO and AI Security Engineer - Skyld AI

Hybrid AI : Why AI has to move from the cloud to On-Device

George Tsirtsis

Sr. Director of Technology - Qualcomm Europe, Inc.

From MLOps to GenOps

Unified data, AI & API marketplaces can drive more synergies and consumption for both data producers and consumers. In this talk I will dive into the successful components of a successful data, AI & API marketplace, and how a platform team can build and run it at scale.

Christophe Lalevée

Google Cloud Platform (Apigee) Customer Engineering Lead - Google

Matthieu Blanc

Machine Learning Product Specialist (South EMEA) at Google Cloud - Google

Bridging Research to Product : AI Innovations into Impact

In this talk, Mehdi Ghissassi, formerly of Google DeepMind and current Chief Product Officer at AI71, explores the journey from research to product in the field of AI and large language models (LLMs). Mehdi explores the challenges and strategies for real-world deployment and highlights the importance of aligning cutting-edge research with market needs, scalability, and user experience.

Mehdi Ghissassi

Chief Product Officer - AI71

Beyond GenAI : Deeplearning for Everyday Usecases

Generative AI (GenAI), with its impressive tools for creating text, images, or sound, has become ubiquitous and attracts all the attention thanks to its visible and accessible results. However, Deep Learning, the technological pillar of GenAI, offers much more than creative tools. In this talk, I will share my experience as a doctoral student at Sonio, where I work on detecting rare fetal malformations, thus aiding in medical diagnosis. Through this lens, I will address how AI and Deep Learning, beyond GenAI, can transform critical fields such as health, the environment, and many other sectors. We will explore how these emerging technologies, although less publicized, can have a profound impact on solving concrete everyday problems. My goal is to show that GenAI is only one facet of AI's potential, and that other fascinating applications are within reach, ready to change our lives in tangible and lasting ways

Hanna Mergui

AI Engineer - Sonio

Cadillacs, Condos and Cardigans : Leveraging AI for Consumer Merchandising

Gaétan Rougevin-Baville

CEO - Meero

Scaling Multilingual Success : Best Practices For Creating Global Websites

This session will be exploring how the integration of language APIs can propel businesses into international markets with unparalleled accuracy and efficiency. Discover how to create seamless localization for websites, apps, and products, without sacrificing customization or data security. Learn practical solutions and get insights into starting with the API - ideal for developers and businesses aiming to break language barriers effectively.

Paul Gordon

Product Marketing, API & Partnerships - DeepL

Floran Pagliai

Head of Engineering - Weglot

De l’impact des imaginaires sur notre relation à la technologie et à l’IA

Une table ronde de 14h à 15h15.

Damien Douani

Educational Innovation, AI, Special Projects at Narratiiv - Podcaster and Presenter of “Chat G Planté, the Price of AI”

Joséphine Corcoral

Secrétaire Générale Adjointe - Conseil National du Numerique (CNNum)

Anne Alombert

Philosophe, membre du Conseil National du Numérique (CNNum)

Hubert Guillaud

Délégué Général - Dans les Algorithmes

Kevin Echraghi

Cofondateur - d’Hérétique

Scaling AI for Computer Vision and Image Generation

As AI adoption in computer vision and generative image models accelerates, the need for efficient deployment solutions on scalable cloud platforms like AWS and GCP becomes critical. In this talk, we will explore how Ikomia SCALE, a powerful SaaS platform, simplifies the deployment of AI models and the effortless creation of APIs for computer vision and generative image tasks. Attendees will learn how our platform enables fast, reliable, and scalable AI implementation, empowering businesses to focus on innovation while seamlessly managing cloud resources and APIs for their AI-driven solutions.

Guillaume Demarcq

CEO & Co-Founder - Ikomia

Tips to reduce the pain of image labeling

In this talk, I will shed some light on one of the main bottlenecks to build successful AI projects: data labeling.
After explaining why this task can be so challenging, I'll share tips to make your labeling processes more efficient
with a focus on Computer Vision projects.Those tips will be divided in three groups. One related to tooling, another
about methodological aspects and the final one about modeling strategies that can help accelerating image labeling.
You'll see that, in some cases, Vision Language Models will be very helpful !

Sonia Tabti

Artificial Intelligence and Computer Vision Specialist, Ph.D.

Data Taggants : Detecting Datasets Usage through Neural Networks

As more machine learning models are deployed without clear information about the data they were trained on, concerns are growing about the misuse of public datasets and the potential for data contamination.
We introduce ""data taggants"", a novel technique that allows dataset owners to verify if their data was used to train a given model with black-box access to the model only.

Wassim Bouaziz

Machine Learning Research - Meta/École Polytechnique

CORE a new era of AI Vision with Vision Language Model (VLM)

William Eldin

CEO - XXII

Building AI Assistants That Work At Scale : Dust's Approach

Gabriel Hubert

CEO and Co-Founder - Dust

The future of Robot Assistants

Jérôme Monceaux

CEO - Enchanted Tools

Build and scale an AI-product in a SaaS solution

Fanny Guélin

Senior Product Manager - Mirakl

Arthur Delaitre

Data Science Manager - AI Catalog - Mirakl

The future of generative AI: exploring new paradigms

Katia Gil Guzman

Developer Experience - OpenAI

AI Evaluation as a Tool for Regulatory Compliance

AI-specific regulations are gaining momentum globally. In order to justify safety and control of AI systems, regulatory provisions increasingly ask for testing and evaluation of AI systems. Among the latter, general purpose AI models are one of the main targets. How do we measure the degree of compliance of an AI system and convey results in an intelligible way to authorities? Evaluation stems as a tool enabling understanding of AI and informed deployment based on regulations. However, it does not come without challenges.

Carlos Muñoz Ferrandis

Creator of OpenRAIL - Co-Founder & COO at Alinia AI

Shadow Compliance for AI

François-Xavier Cao

Co-Founder of SpartaAI - AI Legal services at Olympe.legal

AI Ethics, Trust, and Compliance

A Panel Discussion

Lê Nguyên Hoang

Co-Founder & CEO - Calicarpa

Guillaume Champeau

Legal Director and DPO - Olympe AI

Ramya Chandrasekhar

AI Researcher - CNRS

Jessica Galissaire

Studies and Partnerships Manager - Renaissance Numérique

Make Your LLM Infrastructure Serverless

LLMs are today very popular and very resource consuming, especially in cost expensive ones: the GPUs.

Having a LLM infrastructure to serve different models can be expensive, especially if it is under utilized, and frustrating for the users if it is under sized.

In this session, you will discover how Cloud Run, the serverless flagship service of Google Cloud, can scale up and down LLMs infrastructure and serve at scale your use case, and then, scale down, even to 0 when not used to save money.

Guillaume Blaquiere

Group Data Architect - Carrefour

Embeddings : Core Concepts for Developers

A dense but accessible talk about the core concepts of embeddings.

If you had a collection of every kind of animal on earth, from mules to narwhals to goldfish, how would we pick out just the animals that are housepets? And how does AI distinguish between a pack mule and a Moscow Mule?

Embeddings are numerical representations that capture the essential features and relationships of discrete objects, like words or images, in a continuous vector space, enabling tasks such as semantic search, clustering, and recommendations.

This talk introduces specialized terms and breaks them down into simple, everyday examples to help learners relate difficult concepts to familiar things. Then, it gradually builds back up to the specialized terms. This back-and-forth approach deepens understanding. Attendees will gain a clear understanding of how embeddings work, their benefits, and why they are leveraged in machine learning tasks

Jocelyn Matthews

DevRel, Head of Developer Community - Pinecone

AI Assistants Hands-on! Building with Prompts & Agents

Buckle up, we are about to embark on a journey through human-computer interaction. Remember the good old days of graphical interfaces and mice? Hold on to your keyboards because we’re living in the age of the second revolution: natural language interaction.

In this presentation, we will take a simple but hands-on example to illustrate three different ways of building an AI agent: mainly prompts, combining prompts and code, and LangChain agents. We will see the pros and cons of each implementation.

By the end of this talk, you'll be equipped to start your own explorations in agentic AI!

Marie-Alice Blete

AI Engineer at Komodo Health - Digital Expert AI at Collège Numérique France 2030

Smarter AI, Smaller Models and Reinforcement Learning

Julien Petit

Founder - Mighty Nine

Anais Monlong

Venture Principal - IRIS

Café IA – Mikrodystopia – Writing Workshop

Inscription to the Writing Workshop :
https://events.generationaiconf.com/cafe-ia

François Houste

Café IA Mikrodystopies - Host

2025: AI Winter or Summer? The 3 Bets That Will Change Everything

Nicolas Guyon

IA Générative, Formation IA, Podcast, Speaker, Investisseur, Mentor - Comptoir IA

Why the AI + Automation + NoCode maximizes the return on investment of GenAI?

Generative AI is the new grail for business. But its adoption faces many challenges: training, data management, change management, security, use cases, ....
At Alegria.group, we have supported more than 50 companies of all sizes on their journey towards generative AI for all. By combining new AI, automation and NoCode tools, it has never been so easy to achieve the productivity gains promised by these new technologies.

Francis Lelong

CEO - Alegria.tech

How thriving is the AI hub ecosystem

Mehdi Triki

Head of Public and Institutional Relations - Hub FranceIA

AI in autonomous vehicles. DeepLearning challenges. Ethical issues.

Daniel Vahos Mendoza

AI Engineer - Valeo