The Challenge:
Modernization and improvement of Machine Learning infrastructure.
- There wasn’t enough visibility and monitoring of production predictions of ML models.
- No infrastructure for A/B testing ML models and parameters in production.
- The ML model deployment process was neither standardized nor well documented.
The Engagement:
Novacomp provided a Staff Augmentation service to the Machine Learning and Data Science team.
- Novacomp team integrated into the company native team, participating in each of their stand-ups, Sprint planning, and sync meetings.
- ML team collaborated with other Novacomp staff-augmented teams like Data Engineering and Analytics.
- Novacomp team brought expertise in ML ops technologies and best practices, helping design and implement a new state of the art ML architecture using FastAPI, Split.io, Weights and Biases, Docker, and Apache Pulsar.
The Benefit:
- Implemented a micro service ML architecture that allowed parallel deployments of models for AB testing purposes.
- Developed an Event-driven monitoring system for ML predictions using Apache Pulsar.
- Documented the deployment process, allowing the Data Science team to deploy changes on their own.
- Improved ML services latency by refactoring old code and making asynchronous calls.
Key Technologies:
Python
Scikit Learn
XGBoost
Catboost
FastAPI
Flask
Split.io
Weights & Biases
Docker
Apache Pulsar
Data Feeds / Integrations:
Feeds from Python and Java Microservices Architecture, Snowflake Datawarehouse, and Event-based Streaming with Apache Pulsar.
Integrates into a Sigma Dashboard built by the analytics team.
View our articles
