The client ranks among the world's largest firms in professional services, particularly in the domain of auditing. Their cutting-edge platform facilitates seamless and standardized audits on a global scale.
Using the platform, the client’s Audit teams can spend more time on the areas that matter most, viz., to focus on risks and responses to those risks. The news AI module in the platform plays a key role in enabling the auditors, across the geographies, identify potential risks by scanning news articles relevant to their key end- customers, and extract sentiment, virality and other key insights.
Key Challenges
- Need for Processing large volumes of articles
- Current ML pipeline experienced longer process time
- Solution requires Operational transformation using MLOps
Ready to experience?
TALK TO EXPERTS
The Solution
A centralized MLflow based model leveraging Azure Cloud platform and Databricks Lakehouse platform
- Streamlines the model development and testing across teams.
- Deploys models on containers and exposes them using REST APIs
- Implements federated code repositories and model registries with auto-synching.
- Provides high scalability with features like auto-scaling, streaming capabilities, and easy integration with various data sources as source (EventHub, Azure Blob Storage) or Sink (Delta tables, MongoDB).
- Is highly reliable with better processing speed for varied workloads
A robust MLOps implementation automates testing, integration, deployment, and monitoring.
Implemented the solution leveraging Azure Cloud platform and Databricks Lakehouse platform with a centralized MLflow based model registry to realize streaming and distributed processing capabilities.
Benefits
Processing 80,000 articles in 30 minutes with 2X speed.
12X improvement in manual refresh activity
Processing 7M articles in 40-45 hours, against existing 2 weeks.
Enhanced observability of the ML pipelines.
Enabled ML Model Governance.
