All ProjectsML Pipeline
Real-Time Personalisation Engine for E-Commerce Platform
34% uplift in conversion revenue
34%
Conversion revenue uplift
18%
Average order value increase
2M+
Daily recommendation requests
< 50ms
P99 serving latency
The Challenge
A fast-growing e-commerce platform was using basic rule-based product recommendations that showed every user the same bestseller lists, leaving significant conversion revenue on the table.
Our Solution
We built a two-tower neural collaborative filtering model trained on click, purchase, and dwell-time signals. The model serves personalised recommendations via a low-latency microservice with Redis caching, and includes an A/B testing framework for continuous experimentation. Weekly automated retraining pipelines keep the model fresh as product inventory evolves.
The Outcome
A/B testing confirmed a 34% uplift in conversion revenue from personalised recommendations. Average order value increased by 18%. The system processes over 2 million recommendation requests daily with p99 latency under 50ms.
Tech Stack
PythonPyTorchFastAPIRedisPostgreSQLAirflowMLflowAWS SageMakerNext.js
Category
ML PipelineBuild something similar?
Tell us about your project and we will get back to you within 24 hours.
Start a Conversation
