Ahmad Mateen

Senior AI/ML Engineer with 6+ years of end-to-end experience designing, fine-tuning, and shipping production ML systems across healthtech, enterprise SaaS and related IT fields. Deep expertise in LLMs, RAG architectures, transformer fine-tuning, MLOps pipelines, and real-time inference. Proven track record leading cross-functional teams, reducing model latency at scale, and translating research into revenue-impacting products on AWS.

Lead Machine Learning Engineer at Hopper (2022-04 – Present)

• Spearheaded LLM fine-tuning (GPT-4, Llama 3) and RAG pipeline development, cutting hallucination rate and improving enterprise search relevance scores to 0.91 NDCG.
• Hired via third party to architect an event-driven MLOps platform with MLflow, Kubeflow Pipelines, and GitHub Actions, slashing model release cycles from 3 weeks to 4 days.
• Built real-time NLP microservices (FastAPI + Kafka) handling 50k+ requests/day with P99 latency under 120 ms, deployed on EKS.
• Led a team of 8 ML engineers; established model card standards, drift monitoring with Evidently AI, and automated retraining triggers.
• Designed a multi-modal recommendation engine combining collaborative filtering and CLIP embeddings, lifting click-through rates by 18%.
• Implemented RLHF alignment workflows and prompt-engineering frameworks that reduced human review time by 41% across 3 product lines.

Senior Machine Learning Engineer at Gradient Health (2020-01 – 2022-03)

• Developed FHIR-compliant ML pipelines ingesting EHR data from hospital systems to train predictive readmission models (AUC 0.87).
• Deployed BERT-based clinical NER system that extracted structured diagnoses and medications from unstructured notes with 94% F1 across 6 specialties.
• Built a vector search layer using Pinecone + LangChain to power a clinical decision-support chatbot, reducing physician query time by 27%.
• Containerized inference services (Docker/K8s on GCP) achieving 99.95% uptime SLA for real-time risk-scoring APIs consumed by hospital partners.
• Designed automated feature stores with Apache Feast and Airflow DAGs, reducing data-prep time for new models by 60%.

Data Scientist / ML Engineer at IBM (Watson Health Division) (2017-12 – 2019-12)

• Built and productionized ensemble models (XGBoost + Random Forest) for claims fraud detection, flagging fraudulent claims annually
• Applied transfer learning with ResNet-50 and EfficientNet for medical imaging classification, achieving 91% accuracy on a 200k-image benchmark
• Designed distributed Spark MLlib pipelines on IBM Cloud for petabyte-scale population health analytics across 3 US state health agencies
• Reduced GPU training time by 22% through mixed-precision training and gradient checkpointing on IBM V100 clusters
• Delivered internal tooling and dashboards (Streamlit + Plotly) adopted by analysts across Watson Health business units

MS in Computer Science – California State University (2015-09 – 2017-10)