Projects & Experience

Contributor to the CAI Platform, a scalable MLOps system.

• Kubernetes Cost Optimization: Designed an engine in Golang to analyze live cluster snapshots and simulate VPA recommendations, achieving 15–40% reduction in active nodes.
• Platform Services: Designed Admin Approval System, Audit Logs, and Notification Service using Redis Streams for performance.
• Cost Center Service: Integrated OpenCost and Prometheus for granular pod-level cost tracking.
• Experiment Tracking: Enhanced Python notebooks to capture metadata and metrics for reproducibility.
• Automated Cleanup: Implemented CRON jobs for dataset/model cleanup.
• Data Modules: Led development of dataset uploads and external integrations (HuggingFace, S3) using FastAPI.

• Autism Spectrum Disorder Classification: Proposed MLP-based model with autoencoder pretraining on fMRI scans (ABIDE-1). Achieved 74.82% accuracy.
• Brain Glioma Subtypes: Developed patch-based weakly supervised method for classifying tumor subtypes using gigapixel histology images.