Sana Khan

About

ML/LLM engineer who has deployed a production LLM-powered assistant for intent classification and query routing (order recommendation/support deflection), combining BERT fine-tuning with an embedding-based retrieval layer and optimizing for low-latency inference. Experienced with end-to-end reliability practices—Airflow-orchestrated ETL, data validation/alerting, MLflow experiment tracking, and iterative improvements driven by user feedback and monitoring.

Experience

Education

Key Strengths

• Built and deployed an LLM-based query understanding + routing assistant for order recommendations
• Handled noisy/variable user input via transformer fine-tuning plus embedding-based retrieval
• Production-focused ML evaluation: precision/recall/F1 plus latency and error-rate monitoring
• Iterative improvement using user feedback, low-confidence cases, and refined labeling/training data
• Designed and maintained Airflow ETL pipelines with retries, alerting, and data-quality validation
• Optimized data processing performance (joins, partitioning, column reduction) to reduce runtime
• Made model trade-offs for real-time latency by switching from larger models to distilled models
• Used MLflow experiment tracking to compare versions and gate deployments