Privacy-aware RAG system that automatically detects and masks
PII (names, emails, IDs) from documents before ingestion into
the vector store, preventing sensitive data leakage through LLM
responses.
Production-grade deep learning operations pipeline covering
experiment tracking with MLflow, Docker containerization,
automated CI/CD deployment workflows, model versioning, and
inference serving via REST API.
Real-time facial micro-expression detection system using
CNN-based classifiers with transfer learning achieving 87%
accuracy on FER-2013. Benchmarked across 3 datasets; RAF-DB
model outperformed baseline by 12%.
A full-stack dashboard for monitoring and managing ML model
inference endpoints, with real-time metrics, alert
configurations, and workflow automation via REST API
integrations.