Articles
2026/03/13 : Software architecture tests: 4 examples to get started
https://blog.octo.com/les-tests-d’architecture–4-exemples-pour-les-adopter.
How many times, as a Tech Lead, have you repeated the same remarks in code review? How many times, as an architect, have your recommendations been ignored? How many times, as an AI-assisted developer, have you had to correct generated code that ignored your guidelines? How many times has a technical limitation (e.g. a variable name that is too long) surprised you in production? And how many times has the promised documentation remained just a forgotten ticket?
If you have answered “often” or “too often” to at least one of these questions, architecture tests should appeal to you!
2024/02/20 MLOps: In the run phase, toil is an enemy to face head-on
https://blog.octo.com/mlops–en-phase-de-run-le-toil-est-un-ennemi-a-regarder-dans-les-yeux
⚠️ If the number of people operating your ML products in the run phase is proportional to the number of ML products in production; if development speed has collapsed at the time of going to production — toil has already paralyzed you without you realizing it.
ℹ️ Toil describes the operational work of running a service in production. It can be repetitive, manual, automatable, it does not change the service provided by the system, and it can grow linearly with the number of users.
🔭 Now that some organizations have passed the milestone of putting many ML models into production, they find themselves paralyzed by this toil.
📄 With Aurélien Massiot, we wrote an article on the OCTO Technology blog. On the menu: raising awareness of what it is, how it paralyzes you, how to measure it, and how to address it.
2023/04/12 Architecture Decision Record: let’s start today
https://blog.octo.com/architecture-decision-record/
Published on the OCTO blog, this article describes the Architecture Decision Record as a powerful tool for facilitating and archiving decision-making. It proposes writing ADRs as-code and integrating as-code diagrams.
2023/04/04 Summary of Benjamin Bayart’s talk – “Under the hood of sovereign cloud”
Summary of Benjamin Bayart’s talk at Duck Conf 2023. Benjamin Bayart proposes to talk about cloud under constraints rather than sovereign cloud. He lists the constraints (regulatory, technical, etc.), enumerates existing solutions and encourages us to overcome our impostor syndrome.
2022/12/06 [MLOps] Monitoring & proactive notification of a ML application
https://blog.octo.com/mlops-monitoring-et-proactive-notification/
Published on the OCTO blog, this article details the origins of errors in a ML application in production, and proposes an approach to create monitoring and alerting probes.
2022/09/20 Integrating security early into Machine Learning delivery
https://blog.octo.com/machine-learning-delivery-integrer-la-securite
Published on the OCTO blog, this article addresses the topic of machine learning model security. It describes concrete examples of attacks that impact your models, and proposes many practices to integrate security from the earliest stages.
2022/05/24 Cynefin x Machine Learning matrix – Going fast to production to minimize the risk of complex systems
https://blog.octo.com/matrice-cynefin-machine-learning-quand-produire
Published on the OCTO blog, this article crosses the conceptual framework proposed by Cynefin and Machine Learning to decide whether to go quickly (or not) to production.
2022/05/10 : What if ML monitoring metrics became features?
https://blog.octo.com/et-si-les-metriques-de-monitoring-de-ml-devenaient-fonctionnalites/
Article written with Touraya El Hasssani and Antoine Moreau
Published on the OCTO blog, this article proposes reusing Data Science monitoring metrics as features.
2022/01/25 : Making the value chain visible in a Machine Learning Delivery project
https://blog.octo.com/rendre-visible-la-chaine-de-valeur-dans-un-projet-de-ml-delivery/
Article written with Guillaume Pivette
Published on the OCTO blog, this article is part of the series “Accelerating the Delivery of Machine Learning Projects” dealing with the application of the Accelerate model in a context that includes Machine Learning. In this article in particular, we propose an approach to make your value stream visible and optimized.
2021/10 : Business and Data Scientists must collaborate to design reliable algorithms
https://en.calameo.com/books/004800386209709439859
Pages 31-32 in the POLITHEIS review, this article aims to clarify the responsibilities of the Data Scientist and the business representative in designing reliable algorithms.
2021/07/26 : Test data management in Machine Learning delivery
https://blog.octo.com/la-gestion-des-donnees-de-tests-en-delivery-de-machine-learnin/
Published on the OCTO blog, this article is part of the series “Accelerating the Delivery of Machine Learning Projects”. In this article in particular, we propose practices that will enable effective testing.
2021/03/02 : Visual management in a Machine Learning Delivery project
https://blog.octo.com/la-gestion-visuelle-dans-un-projet-de-machine-learning-delivery/
Article written with Maria Mokbel
Published on the OCTO blog, this article is part of the series “Accelerating the Delivery of Machine Learning Projects”. In this article in particular, we propose practices for visualising work in progress and quality in a Machine Learning delivery.
2021/01/13 : Our 10 convictions to better succeed in our Data Science projects in 2021
https://blog.octo.com/reussir-ses-projets-de-data-science-en-2021
Article written with Eric Biernat
Published on the OCTO blog, this article proposes 10 orientations we observe at the start of 2021 in the Data Science market. Methodology, organisation, collaboration, and continuous improvement are the themes covered.
2020/05/13 : [MLOps] The challenges of retrieving the ideal prediction
Published on the OCTO blog, this article digs into one of the difficulties in tracking the real performance of a Data Science algorithm: retrieving the prediction ideally made by the algorithm to maximise the achievement of its objectives.
2019/12/17 : [MLOps] An alternative to distribution monitoring
https://blog.octo.com/une-alternative-au-monitoring-de-distributions/
Article written with Mehdi Houacine
Published on the OCTO blog, this article proposes a methodology to obtain meaningful metrics and alerts for a Data Science system. This methodology was also presented at Duck Conf 2020. This article is one of the first French articles on Data Science monitoring.
2019/11/20 : Interpretability of Data Science Systems
https://blog.octo.com/interpretabilite-des-systemes-de-data-science/
Published on the OCTO blog, this article aims to frame the problem of Data Science system interpretability. In 3 parts, we discover what is meant by interpretability (for whom? why? what? and at what effort?), then we explore the reasons for non-interpretability, and finally we propose some solutions.
2017/10/26 : Benchmark of dataPreparation versus standard R
https://rpubs.com/ELToulemonde/326980
Published on RPUBS.com, this article presents the benefits of the dataPreparation library compared to standard R on a classic data preparation operation: centering and scaling data.
2017/09/14 : On data and quality
https://www.linkedin.com/pulse/de-la-donnée-et-qualité-emmanuel-lin-toulemonde
Published on LinkedIn, this article traces the reasons for poor data quality and proposes 5 actions to improve it.
2017/08/31 : AI, the end of employment?
https://www.linkedin.com/pulse/ia-la-fin-de-lemploi-emmanuel-lin-toulemonde
Published on LinkedIn, this post is a reflection on the impact of various technological revolutions on employment, from the agricultural revolution to that of artificial intelligence.