Ahmad Hayes

Designed and deployed end-to-end Generative AI systems using transformer-based LLMs for enterprise NLP use cases including conversational AI, summarization, and semantic search; Built and scaled Retrieval-Augmented Generation pipelines with FAISS and Pinecone for real-time contextual search; Led development of intelligent document understanding with entity extraction and OCR pipelines; Built production-grade ML APIs via FastAPI and containerized services with Docker and Kubernetes; Created automated pipelines for prompt engineering, model evaluation, and multi-run experiment tracking using MLflow and Weights & Biases; Explored LoRA/PEFT for parameter-efficient fine-tuning; Integrated LangChain for modular LLM apps; Established internal A/B testing and latency benchmarking; Used Kineto trace and Torch dispatcher to profile kernels and optimize GPU runtime; Collaborated with MLOps to deploy on AWS/GCP.

Led multilingual NLP systems for sentiment analysis, text classification, and information extraction using BERT and XLNet; Deployed models via REST APIs using Flask and TensorFlow Serving for real-time predictions; Engineered pipelines for time-series forecasting and anomaly detection to monitor KPIs; Built transfer learning workflows to reduce data requirements and improve cross-domain generalization; Implemented automated ETL and model orchestration with Apache Airflow and Docker; Designed modular ML pipelines with feature extraction, training, hyperparameter tuning, and drift-based retraining; Partnered with product and engineering teams to define problems, metrics, and evaluation plans; Implemented data labeling and active learning to boost annotation throughput; Promoted ML best practices and reproducible pipelines.

Led multilingual NLP for sentiment analysis, text classification, and information extraction using modern models; Integrated pre-trained models like BERT and XLNet into customer-facing applications; Engineered pipelines for time-series forecasting and anomaly detection to predict operational KPIs; Developed custom transfer learning workflows to reduce training data requirements and boost generalization across domains; Deployed models via REST APIs using Flask and TensorFlow Serving for real-time predictions; Created automated ETL workflows and model orchestration pipelines using Apache Airflow and Docker for reproducibility and scalability; Designed modular ML pipelines with feature extraction, training, hyperparameter tuning, and automated retraining based on drift detection; Partnered with product and engineering teams to define ML problem statements, success metrics, and iterative experiment cycles; Implemented data labeling and active learning strategies to optimize dataset quality