AlgoDeep
Creating a platform to scale algorithms
How did Alien6 design and implement the foundation to deploy industrial and secure Scientific Artificial Intelligence within the Research and Innovation department of one major cosmetic player?
Services
- Architecture
- Integration
- Data Science
Industry
- Cosmetic
Years
- 2018-2021
Duration (CAPEX)
- 12 months
Our Mission
The team in charge of Scientific Computing wanted to facilitate the diffusion of algorithms and in-house scientific models to all 4000 researchers. Through IT teams, Alien6 had the privilege of being asked in early 2018 to respond to those ambitions, design, implement and safely maintain a platform relying on Cloud capabilities to exhibit new computations at scale within the Information System of the Research & Innovation department. Once exposed, those new capabilities could offer the possibility of designing applications in compliance with architecture and security requirements and standards.
The objective was twofold: first, it aimed to facilitate exchanges between scientific teams and maximize the uses of the Information System. The idea was not only to share Notebooks between analysts but also to build a complete chain of knowledge that can benefit all our researchers by relying on the expertise and know-how of our colleagues. Within the architecture team, we also wanted to take the opportunity of this project to assess our ability to integrate the technologies suggested by Alien6 in our reference architecture and then identify, using a fine mesh, the impacts on our various IT processes.
,
The customer
Pure Player and Leader in Beauty, this century-old and international group in the cosmetics industry relies on the excellence of its employees and partners to satisfy the plurality of needs and aspirations of each in terms of Beauty. In a process of continuous improvement, the Research & Innovation teams strive to be able to offer everyone, always a beauty more inclusive and personalized based on ever more precise use of data, algorithms and Artificial Intelligence.
Requirements & Constraints
Plurality
Algorithms can be of different types (Statistics, Machine Learning, Deep Learning, ...), whatever the language used (R, Python, C / C ++, ...) and whatever the library (Keras, TensorFlow, ...)
Elasticity and Adaptativity
Whether they require a few milliseconds or several days of processing, algorithms must be able to be called on demand and by anyone with all appropriate authorizations.
Hybridization
The platform must be able to serve algorithms distributed locally or on a public cloud in compliance with the group's safety standards.
Compliance and Supportability
The platform must be installed, maintained and supervised by the teams in place.
Users
Data Scientists
Data Scientistsare responsible for designing models and algorithms which will be consumed by various software components once deployed on the platform.
Researchers
Like Data Scientists, Researchers exhibit their models on the platform. Others rely on a library of algorithms or applications that integrate libraries to validate their working hypotheses.
Developers
Developers rely on augmented services to design smart applications for researchers or to extend the capabilities of the existing Information System.
Results
An augmented and smart software chain
Models can be consumed simultaneously by dozens of applications and indirectly by hundreds of researchers. Data Scientists can refine their models and regularly update them without worrying about the underlying technical constraints.