At the end of November 2024, I proudly joined a company whose members I've admired for over 15 years, since the day I wrote import sklearn which shaped my career in Data Science. My role is twofold: developing the Data Science consulting team and collaborating with users and paying customers, deepening our understanding of industrial needs while nurturing our Open Source cause.

Laurent Pantanacce and I applied our heedgi methodology to help glasspop meet its ambitious challenge. Our approach reduced the number of optical tests needed per patient through the integration of proprietary artificial intelligence that optimizes patients' optical comfort.

This work was presented at two specialized ophthalmology conferences:

• Refractive Surgery & Innovations
  January 20, 2024


• Imaging in Ophthalmology
  October 4, 2024

Our involvement covered many aspects of the project, from technological innovation to scientific communication and strategic business negotiations. This collaboration illustrates the potential of combining science and medical practice for patient benefit, redefining eyewear prescription for years to come.

What are Volpi's potential goals with AI?

Reduce costs through AI in software within hardware
‣ Amplify medical annotation efforts with a human-in-the-loop approach
‣ Guide and advise from R&D to production avoiding costly pitfalls

Through the heedgi methodology, we focused our attention on a clinical problem and how best to measure success: the unit of value and KPI to direct, choose and build our custom AI for Volpi.

Acculturation, advanced technical advice, and clinical understanding were key elements of the custom program we conducted for Volpi. Our goal was to prepare Volpi for a fast track to client success. Having our heedgi methodology help establish a solid foundation for AI at Volpi, from management to technicians, scientists, and salespeople, is a source of satisfaction.

I asked my manager, Nicolas Renard, if I could quote this phrase 👆, and he agreed. Thank you, Nicolas, not only for being a great manager but also for your incredible irreplaceable substitution of the MBA course at Rennes School of Business! 🙏🙏🙏

At Jellysmack, my team went all out to help the web video creation space with innovative projects in computer vision and natural language processing serving creators worldwide.

The mission was twofold:
‣ Foster innovation through prototypes
‣ Provide research support to other AI/ML teams by bringing fresh, bibliographic, and pragmatic insights

We strive to invent and improve secret sauces, diligently craft deliverable code, conduct scientific and technological watch, and participate in internal and external discussions.

The greatest pride?

The video representation project, a combination of multiple kernel learning and Nyström approximation.

We meticulously gathered and harmonized the best video representations from research. Through our media experts' invaluable knowledge, we not only inferred our own combined similarity but also designed fixed-size vector representations for efficient clustering, retrieval, and success prediction estimation.

We changed the video editing profession by leveraging the work of our hundreds of professional editors, optimizing quick editing and short conversion processes for platforms such as YouTube, Facebook, Instagram, and TikTok. By analyzing multiple high-level signals and designing various cutting methods, we provide suggestions to editors, ensuring a seamless and creative editing experience, minimizing time spent on tedious, non-artistic tasks.

From a scientific standpoint, it's intellectually challenging to process videos as this medium is multimodal: image, sound, speech, music, and movement. On the business side, these areas are very rewarding when considering video success and reduced time spent building them, clearly defining our pragmatic performance indicators.

This course bridges the gap between AI academic experts and business leaders. For managers, business leaders, and technology enthusiasts, this course is designed as a beacon guiding professionals through the maze of AI fundamentals, project management, and real-world application, avoiding pitfalls in a sustainable business.

With over 200 slides of constantly updated and tested material, we distill refined knowledge from case studies, practical examples, and interactive sessions. The program, which has been deployed more than three times, is a tapestry weaving AI basics, challenges, cutting-edge technologies, and effective strategies, all aimed at empowering participants to lead AI projects with confidence and even subtlety, benefiting customers and sustainable businesses.

With a Ph.D. in applied mathematics and a seasoned tech entrepreneur as instructors, the course is more than a leap of faith; it's a jump towards AI mastery and business success.

The goal of Ircam Amplify is to blend technology, mathematics, and music, creating harmonious solutions that resonate with science and music while striking a chord in the market. In this mission, we sought to maximize profits on Spotify from our music catalogs.

This involved creating innovative algorithms to accurately analyze, encode, and price both old and new tracks. We leverage data dynamics, guiding music enthusiasts on a personalized discovery journey, while considering latest trends and listeners' success history.

By intelligently clustering tracks and adaptively encoding them, and leveraging Ircam's vast research capital, we enable AI systems to appreciate music. We also successfully experimented with connecting the news feed to improve suggestions.

In four life-changing years at MAP5 (2016-2020), the applied mathematics laboratory of the University of Paris, I immersed myself in the exciting field of generative and discriminative AI. Taking a close interest in neural networks and deep learning, I repurposed these powerful tools. Notably, I developed clustering techniques with GANs, enabling the organization of vast datasets like large photo collections. I also explored the importance of unsupervised features, revealing distinctive attributes of unlabeled datasets.

A happy coincidence allowed me to meet Dr. Élodie Germani during an internship I was fortunate to supervise with Dr. Alkéos Michaïl. Élodie is now a researcher specializing in artificial intelligence for medical imaging.

In parallel, my academic and professional work focused on methods for evaluating the reliability of automated decisions, providing an essential statistical approach for applications requiring human supervision. This exploration merged certain optimal transport mathematics with neural networks, thus allowing data representation to be reassessed. Thanks to MAP5's human warmth and perseverance at Oscaro, my journey has modestly contributed to a field that I hold very dear.

Bringing back publications from top-tier research conferences and journals provides a feeling of pride, vanity, and recognition, but also, of course, the unwavering will to fight the next battle of research in business.

At Oscaro.com, I participated in innovative initiatives as an R&D Data Scientist, seamlessly combining deep learning and artificial intelligence to develop robust solutions. My work was instrumental in implementing cutting-edge mechanisms for credit card fraud detection, significantly reducing risks and protecting transactions. In parallel, I focused on demand forecasting, using deep learning to accurately predict market dynamics, thus improving efficiency and customer satisfaction.

Credit card fraud detection was a good example of the importance of asking the right questions. At the time, companies were content with counting errors, but my approach was to assess the company's financial losses. All AI systems, like others, make mistakes: my role is to handle them in the most meaningful way possible. This approach radically changes how algorithm performance is measured, rewarded, and therefore trained. It also benefits management, as the entire chain of command, even without AI expertise, now understands that the main goal is to protect company money. This perspective has facilitated my work with companies and clients.

My role also included AI technology watch, where I monitored and analyzed emerging trends, research, and companies in the AI field, ensuring Oscaro.com stayed at the forefront in a competitive landscape. This commitment to excellence also materialized through the pursuit of my doctorate, generously funded from September 2016 to October 2020. Moving from university to business enriched my vision of AI, nurturing both my intellectual thirst and professional expertise.

Within the Oscaro.com group was the startup xBrain pioneering in chatbots and I had the pleasure of supervising an internship for Dr. Maxime Haddouche now a researcher in Machine Learning and numerical optimization.