Maciej Szmida

Architected production ML inference pipelines for multimodal AI models, reduced latency by 40% through model quantization and optimized batching with Docker and Kubernetes. Deployed fine-tuned LLMs for content classification and retrieval using PyTorch and Transformers, achieving 95% accuracy on AB tests and supporting 500K daily user requests. Implemented RLHF-based retrieval augmented generation with FAISS vectors search, improving content discovery relevance by 32% and reducing cold-start recommendations. Designed CI/CD for ML pipelines using MLFlow and GitHub Actions, cutting deployment time from hours to 15 minutes and enabling reproducible experiments. Built real-time model monitoring dashboards with Prometheus and Grafana, detecting data drift within 5 minutes and maintaining 99.9% uptime. Optimized GPU utilization on AWS SageMaker and EKS, reducing inference cost per request by 28% while increasing throughput by 3x for video processing pipelines. Integrated AI chat bot for inte

Designed and developed AI agents to automate client work processes; implemented predictive models and data-driven capabilities; led end-to-end ML lifecycle from data processing to model deployment. Collaborated with cross-functional teams to translate requirements into scalable AI solutions.

Engineered ML platform; built event-driven data pipelines with Kafka and Apache Airflow, processing 10 TB+ daily telemetry data and feeding real-time features to inference models. Designed and deployed RESTful APIs for model services using FastAPI and gRPC, handling 10,000+ requests per second with 99.99% uptime. Implemented CI/CD for ML pipelines with MLFlow and GitHub Actions, enabling reproducible experiments and reducing deployment time. Built real-time model monitoring dashboards with Prometheus and Grafana, detecting data drift within 5 minutes and maintaining 99.9% uptime. Optimized GPU utilization on AWS SageMaker and EKS, improving throughput and reducing costs. Integrated AI chatbots for internal knowledge retrieval and data-driven decision making.