Sreeja N

Ai Ml Engineer @Scale AI

San Francisco, CA, US

Designed and automated large-scale data-curation and evaluation pipelines with Python, Airflow, and FastAPI, improving labeling throughput by 65 % and enabling enterprise-grade dataset processing.Engineered Quantization-Aware Training (QAT) and model-compression workflows using PyTorch and ONNX Runtime, cutting inference cost by 40 % while maintaining accuracy across production LLMs.Built scalable MLOps infrastructure with MLflow, DVC, and AWS EKS for reproducible experiments, CI/CD automation, and rollback-safe deployments across multi-tenant model pipelines.Developed evaluation frameworks leveraging Transformers, scikit-learn, and Spark SQL to measure coherence, factual accuracy, and bias over multilingual datasets (> 2 billion samples).Created interactive Tableau and Power BI dashboards for tracking quantization impact, bias metrics, and dataset drift, reducing audit and retraining cycles by 35 %.Contributed to Meta’s LLaMA 3.1 (405 B) initiatives through Scale AI’s enterprise collaboration, focusing on dataset alignment, QAT, and ONNX optimizations for efficient large-model deployment.Optimized edge-inference pipelines via FastAPI, TorchServe, and AWS SageMaker, achieving p99 latency < 500 ms on quantized LLaMA variants (8 B / 13 B).Partnered with Meta’s AI Infrastructure and Responsible AI teams to ensure GDPR-compliant data exchange, implement bias detection, and maintain transparent model-evaluation governance.

Jawaharlal Nehru Technological University

Bachelors, Computer Science

California State University - East Bay

Master's degree, Computer Science