Amara Nasir

I’m a Machine Learning Engineer with 5+ years building production ML and GenerativeAI systems across fintech fraud detection and large-scale document intelligence. I focus on delivering end-to-end pipelines that reliably turn raw data into dependable model outputs, not just experiments.

On Azure Databricks and Azure Data Factory, I design and orchestrate LLM/RAG pipelines for document understanding, extraction, and classification at scale. I route each stage to the best-fit model to balance accuracy, latency, and cost, and I engineer high-throughput asynchronous workflows with structured-output validation (Pydantic) and automatic retries for schema-conformant results.

I also build the data foundations behind these systems: scalable PySpark flows over JDBC to relational databases, data cleaning and fuzzy matching across large datasets, and fault-tolerant I/O with ADLS Gen2 while managing credentials securely through Azure Key Vault. My work includes ground-truth evaluation frameworks—metrics, confusion matrices, and confidence-threshold analysis—to benchmark performance and continuously improve prompts and routing logic.

Previously, I engineered fraud-detection models on millions of transactions and improved model performance by 50% while increasing fraud detection by 6%. I bring a strong full-stack and data-engineering foundation—Python services, REST APIs with FastAPI, and responsive web application development—so models integrate smoothly into real products and production systems.