Nadia Ayaz

Architected agentic AI solutions across more than 13 enterprise workflows using Google Cloud, Vertex AI, LangChain, and vector search. Built a multi-agent social media content platform coordinating seven or more specialized agents through Pub/Sub and MCP-based tool calling. Designed Firebase and BigQuery data and agent-state architecture, improving document retrieval accuracy by 13% and reducing manual review time by more than 29%. Established LLM evaluation and observability frameworks using LangSmith and custom scoring across more than 26 prompt variants.

Built large-scale AWS data pipelines with Glue and Redshift for real estate lead data, reducing pipeline runtime by 31% and query costs by 23%. Architected an anomaly detection solution on SageMaker using Docker that processed more than 10,000 lead records monthly and improved model accuracy by 29%. Led Airflow ETL and ELT orchestration across more than 15 data sources and integrated SageMaker inference into lead-scoring workflows. Developed a WhatsApp appointment scheduling service using Amazon Pinpoint, Lambda, and the WhatsApp Business API.

Architected distributed NLP and document intelligence pipelines using Apache Spark and PySpark to transform millions of fragmented legal case files into structured, searchable knowledge. Delivered FastAPI microservices for semantic search and retrieval, enabling natural-language access to legal histories. Built automated demand-letter generation by combining structured case data with retrieved medical and incident reports, reducing manual drafting effort by approximately 40%. Developed fraud detection and risk-scoring models using gradient-boosted trees and anomaly detection.