ML | ClaimWise AI
About the project
As a team, we all experienced the frustration of slow claims processing and lengthy adjustments when dealing with insurance. This inspired us to build a scalable fraud detection workflow that automates claim evaluation, identifies potential fraudulent activity, and speeds up decision-making. Our goal was to create a workflow that could handle large datasets efficiently while providing actionable insights.
Project Scope / Question
How can multiple data modalities be combined to improve fraud detection?
Which features are most predictive of fraudulent claims?
How can we scale and automate scoring with BigQuery AI?
Approach / What I Did (My Contributions):
Integrated structured and semi-structured claim data into a unified feature set
Performed data cleaning, feature engineering, and preprocessing
Trained and evaluated machine learning models for anomaly detection
Leveraged BigQuery AI to automate large-scale scoring and inference
Coordinated with team members to align modeling and feature selection
Impact & Innovation
ClaimWise demonstrates production-ready AI that transforms insurance operations through native BigQuery ML processing. Our multimodal approach combines text embeddings, image analysis, and vector similarity search to deliver sub-30-second claim processing with built-in fraud detection.
Technical Achievement: Seamless integration of ML.GENERATE, ML.GENERATEEMBEDDING, and VECTORSEARCH functions eliminates data movement overhead while scaling automatically.
Business Impact: 95% faster processing, 40% manual effort reduction, and instant and proactive fraud detection—proving that intelligent automation can revolutionize traditional insurance workflows.
Tools used
Python (Pandas, scikit-learn)
BigQuery AI
SQL
Jupyter Notebook
Check out our Kaggle writeup: