Jelmer Borst, analytics and machine learning leader at Picnic, and Daniela Solis Morales, machine learning lead, delve into the dynamics of building effective ML teams. They discuss shifting from decentralized to centralized structures and the challenges of recruiting the right talent. The pair explores the complexities of demand forecasting in online grocery delivery and stresses the importance of collaboration between data scientists and business teams. They also highlight the need for lightweight, scalable ML infrastructure and the evolving roles within data science to meet business goals.
Read more
AI Summary
AI Chapters
Episode notes
auto_awesome
Podcast summary created with Snipd AI
Quick takeaways
Picnic transitioned from a decentralized to a centralized machine learning team, enhancing collaboration and efficiency in their projects.
The importance of a centralized data engineering team at Picnic significantly improved access to clean data, accelerating model development processes.
Picnic fosters a strong feedback loop between machine learning models and business outcomes, ensuring practical applications align with business goals and market demands.
Deep dives
The Journey from Decentralized to Centralized Machine Learning
The machine learning team at Picnic transitioned from a decentralized to a centralized approach, contrasting with the experience of Delivery Hero, which started with a centralized model and later decentralized. Initially, Picnic's machine learning engineers faced challenges due to the decentralized model, especially regarding knowledge sharing and maintaining best practices. After recognizing the benefits of centralization, such as improved collaboration and efficiency, the team decided to consolidate efforts to enhance their overall machine learning capabilities. This restructuring allowed for quicker iterations on machine learning solutions and helped the team manage growing demands effectively.
Leveraging Clean Data for Machine Learning Success
A key factor contributing to Picnic's success in machine learning is the access to clean, well-structured data, established through a centralized data engineering team. This centralization minimized duplication of effort and allowed machine learning engineers to focus on model creation rather than data preparation, significantly speeding up the development process. With the team building on top of a robust data infrastructure that included Snowflake, they were able to extract insights rapidly. This clean data foundation helped the team experiment and iterate on machine learning models without the common bottleneck of data quality issues.
Building a Feedback Loop for Model Performance
Picnic emphasizes a strong feedback loop to evaluate the impact of machine learning models on business outcomes, particularly in operational efficiency. This involves monitoring not just the performance metrics of the models but also their actual business value, such as product availability and waste reduction. By aligning the goals of machine learning engineers with those of business analysts, the team ensures that models are designed to deliver meaningful results. Continuous evaluations help refine the models and adapt to shifting market demands, maintaining a focus on practical application.
Importance of Scalability in Model Management
As Picnic's machine learning efforts expanded, the management of multiple models became increasingly complex, leading to concerns about tracking and maintaining model performance. To address this, the team implemented tiered monitoring strategies to distinguish between models in various phases of experimentation and those that are operationally critical. They also focused on fostering collaboration between technical and business teams to ensure alignment and transparency regarding model impacts. By prioritizing effective communication and shared objectives, the team mitigates risks associated with deploying numerous models.
Strategic Team Organization for Efficiency
The structure of the machine learning teams at Picnic plays a crucial role in enhancing efficiency and innovation. With 18 data scientists organized into specialized teams focused on operational and consumer-related tasks, the structure supports both collaboration and specialization. While some projects benefit from embedding machine learning engineers within product teams, the centralized approach has proven effective for scaling efforts without sacrificing knowledge sharing. This strategic organization aligns the team's skills with business needs, ensuring that machine learning solutions can grow alongside the company's objectives.
The AI Dream Team: Strategies for ML Recruitment and Growth // MLOps Podcast #267 with Jelmer Borst, Analytics & Machine Learning Domain Lead, and Daniela Solis, Machine Learning Product Owner, of Picnic.
// Abstract
Like many companies, Picnic started out with a small, central data science team. As this grows larger, focussing on more complex models, it questions the skillsets & organisational set up.
Use an ML platform, or build ourselves?
A central team vs. embedded?
Hire data scientists vs. ML engineers vs. MLOps engineers
How to foster a team culture of end-to-end ownership
How to balance short-term & long-term impact
// Bio
Jelmer Borst
Jelmer leads the analytics & machine learning teams at Picnic, an app-only online groceries company based in The Netherlands. Whilst his background is in aerospace engineering, he was looking for something faster-paced and found that at Picnic. He loves the intersection of solving business challenges using technology & data. In his free time loves to cook food and tinker with the latest AI developments.
Daniela Solis Morales
As a Machine Learning Lead at Picnic, I am responsible for ensuring the success of end-to-end Machine Learning systems. My work involves bringing models into production across various domains, including Personalization, Fraud Detection, and Natural Language Processing.
// MLOps Jobs board
https://mlops.pallet.xyz/jobs
// MLOps Swag/Merch
https://mlops-community.myshopify.com/
// Related Links
Website:
--------------- ✌️Connect With Us ✌️ -------------
Join our slack community: https://go.mlops.community/slack
Follow us on Twitter: @mlopscommunity
Sign up for the next meetup: https://go.mlops.community/register
Catch all episodes, blogs, newsletters, and more: https://mlops.community/
Connect with Demetrios on LinkedIn: https://www.linkedin.com/in/dpbrinkm/
Connect with Jelmer on LinkedIn: https://www.linkedin.com/in/japborst
Connect with Daniela on LinkedIn: https://www.linkedin.com/in/daniela-solis-morales/
Get the Snipd podcast app
Unlock the knowledge in podcasts with the podcast player of the future.
AI-powered podcast player
Listen to all your favourite podcasts with AI-powered features
Discover highlights
Listen to the best highlights from the podcasts you love and dive into the full episode
Save any moment
Hear something you like? Tap your headphones to save it with AI-generated key takeaways
Share & Export
Send highlights to Twitter, WhatsApp or export them to Notion, Readwise & more
AI-powered podcast player
Listen to all your favourite podcasts with AI-powered features
Discover highlights
Listen to the best highlights from the podcasts you love and dive into the full episode