WeAreDevelopers Logo WeAreDevelopers
Sign in
For You LIVE
World Congress 2023 • Oct 6, 2023

How We Built a Machine Learning-Based Recommendation System (And Survived to Tell the Tale)

Dora Petrella

A data science team replaced their underperforming SQL recommender with a Meta-Prod2Vec neural network. Learn how injecting article metadata solved long-tail recommendations and boosted product coverage by 45 percent.

Pause
Mute Enter Fullscreen
#1 about 2 min

Identifying the need for product replacement recommendations

Out-of-stock items drive customers to competitors unless relevant alternatives are immediately surfaceable.

#2 about 2 min

Understanding data sources in a complex b2b environment

Combining internal systems, website events, and purchase histories provides the foundational data for machine learning models.

#3 about 3 min

Defining multiple business use cases for recommendations

A flexible recommendation architecture serves diverse needs ranging from customer-facing web pages to internal order management tools.

#4 about 3 min

Evaluating limitations of legacy sql-based recommenders

Statistical formulas calculated continuously in sql struggle to provide accurate recommendations for newly listed or unpopular items.

#5 about 3 min

Adapting natural language algorithms for product recommendations

Applying prod2vec logic to user interaction sessions allows engineers to derive implicit training labels using sliding context windows.

#6 about 2 min

Training a neural network with one-hot encoded vectors

Feeding product combinations into a hidden layer model generates probability scores for items appearing within the same session.

#7 about 2 min

Improving long-tail predictions using meta-prod2vec algorithms

Concatenating category and brand metadata into product vectors significantly improves alternative matching for infrequently purchased items.

#8 about 1 min

Calculating vector distances to generate final alternatives

Measuring cosine similarity between geometric array representations allows the system to surface the most contextually relevant product replacements.

#9 about 2 min

Building the machine learning engineering pipeline

Connecting dbt transformations with vertex ai and orchestrating regular automated training runs ensures clean data flows into the elasticsearch storage.

#10 about 3 min

Evaluating model performance through offline metrics mapping

Testing against a production-like distribution utilizing ndcg and shadow traffic analysis confirms improvements in latency and catalog coverage.

#11 about 4 min

Visualizing product embeddings to validate clustering logic

Plotting multidimensional vectors in an open-source 3d projector tool confirms that related product attributes cluster accurately.

#12 about 2 min

Establishing pragmatic business baselines for model evaluation

Implementing a simple heuristic baseline metric sets a measurable quality gate and builds trust with non-technical stakeholders.

#13 about 2 min

Automating exploratory data analysis within training pipelines

Embedding automated html plotting directly into the model training pipeline prevents bad data from silently corrupting updated training runs.

#14 about 2 min

Team composition and realistic product launch timelines

Delivering an end-to-end recommendation system requires cross-functional collaboration and often exceeds initial baseline timeline estimates.

Matching moments

1:50 min

Real life recommendation systems and final project conclusions

Lutske van der Meer Lutske van der Meer · World Congress 2024

2:40 min

Understanding real-world recommendation systems in common platforms

Julian Joseph · LIVE

2:14 min

Assessing the future of AI in web performance optimization

Perf + AI

4:42 min

Predicting future retail experiences and avoiding technology hype

Alejandro Saucedo Alejandro Saucedo +2 · World Congress 2025

57 sec

Implementing predictive prefetch with machine learning

Jessica Janiuk · JS Congress

2:50 min

Driving subtle product enhancements with historical platform data

Alejandro Saucedo Alejandro Saucedo +2 · World Congress 2025

More from World Congress 2023

The year 3000, a brief history of Web Development thumbnail
25:30
over 2 years ago

The year 3000, a brief history of Web Development

Lorenzo Pieri

Demografie nutzen um jedem Bewerber das passende Angebot zu machen! thumbnail
8:52
over 2 years ago

Demografie nutzen um jedem Bewerber das passende Angebot zu machen!

Nathalie Gierth

Guided Memory Management: Rust's Ownership Model thumbnail
0:18
almost 3 years ago

Guided Memory Management: Rust's Ownership Model

Stefan Baumgartner

Passwordless future: WebAuthn and Passkeys in practice thumbnail
32:32
almost 3 years ago

Passwordless future: WebAuthn and Passkeys in practice

Clemens Hübner

External Secrets Operator: the secrets management toolbox for self-sufficient teams thumbnail
30:07
almost 3 years ago

External Secrets Operator: the secrets management toolbox for self-sufficient teams

Moritz Johner

Building Reliable Serverless Applications with AWS CDK and Testing thumbnail
25:18
almost 3 years ago

Building Reliable Serverless Applications with AWS CDK and Testing

Raphael Manke

Upcoming sessions on this topic

Open session

World Congress 2026 North America

September 25, 2026 · 16:50–17:20

Stage 4

Engineering Moneyball: How We Benchmarked Google vs Meta

Jirka Bachel

Co-Founder & CEO

Jirka Bachel
Open session

World Congress 2026 North America

September 25, 2026 · 16:10–16:40

Stage 1

The Five Percent Club: The Culture and Technological Shift Behind Successful AI Deployments

Tara Hernandez

Tara Hernandez, VP of Developer Productivity at MongoDB

Tara Hernandez
Open session

World Congress 2026 North America

September 25, 2026 · 09:00–09:30

Outdoor Stage

Engineering the Pivot: How Creative Strategy Solves the Hard Problems of AI Accuracy and Scale

Shruti Tiwari

AI/ML product manager, Dell

Shruti Tiwari
Open session

World Congress 2026 North America

September 24, 2026 · 16:50–17:20

Mainstage

Building AI Products vs. Building With AI

Aparna Dhinakaran, Rukmini Reddy, Tamar Bercovici

Aparna Dhinakaran
Rukmini Reddy
Tamar Bercovici
Open session

World Congress 2026 North America

September 25, 2026 · 09:00–09:30

Mainstage

Building AI that fits your business

Benny Chen

Co-Founder of Fireworks

Benny Chen
Open session

World Congress 2026 North America

September 25, 2026 · 11:40–12:10

Stage 9

You Can’t Re-Run Sunlight: Designing ML Data Architectures for Physical AI

An Phan

Senior Data Infrastructure Engineer @ Hippo Harvest

An Phan

Related videos

Data Science in Retail thumbnail
58:21
LIVE • over 3 years ago

Data Science in Retail

Julian Joseph

What non-automotive Machine Learning projects can learn from automotive Machine Learning projects thumbnail
48:01
World Congress 2022 • about 4 years ago

What non-automotive Machine Learning projects can learn from automotive Machine Learning projects

Jan Zawadzki

Is my AI alive but brain-dead? How monitoring can tell you if your machine learning stack is still performing thumbnail
48:07
LIVE • almost 5 years ago

Is my AI alive but brain-dead? How monitoring can tell you if your machine learning stack is still performing

Lina Weichbrodt

Building Products in the era of GenAI thumbnail
59:39
LIVE • almost 3 years ago

Building Products in the era of GenAI

Julian Joseph

Deployed ML models need your feedback too thumbnail
37:24
World Congress 2021 • about 5 years ago

Deployed ML models need your feedback too

David Mosen

Anomaly Detection - Using unsupervised Machine Learning for detecting anomalies in customer base thumbnail
22:06
LIVE • about 6 years ago

Anomaly Detection - Using unsupervised Machine Learning for detecting anomalies in customer base

Lukas Kölbl

Related articles

From learning to earning

Jobs that call for the skills explored in this talk.