Data Scientist

Large Language Models (LLMs), ChatGPT, Prompt Engineering, Retrieval Augmented Generation (RAG), Chatbots, NLP, Keyword Extraction, Keyword Matching, Semantic Search, LangChain, FastAPI.

• Implementing Time Series Forecasting models to predict the power generation of thousands of renewable plants for both intra-day and day-ahead markets using LSTM, XGBoost, LightGBM, and Random Forest.
• Implementing Anomaly Detection algorithms to find extreme values in Time Series Data.
• Conducting extensive data preprocessing and feature engineering to improve models' accuracy and scalability.
• Designing Machine Learning development and production environments with microservice architecture in Kubernetes clusters and Kubeflow.
• Deploying machine learning models into production using various deployment strategies, such as REST APIs, KServe, BentoML, and containerization technologies like Docker and Kubernetes.
• Continuously monitoring and training deployed machine learning models using performance metrics and user feedback, making adjustments and improvements as necessary to meet business requirements.
• Developing and maintaining core microservices for the production environment using FastAPI, Flask, and Node.js.
• Proficient in BERT, LLM, ChatGPT, Fine-Tuning, NLP, LLaMa, RAG, ChatGPT API, LangChain, and Agents.

• Processed and analyzed high-dimensional 3D TOF MRA images under the supervision of a radiology group.
• Developed, optimized, and implemented 3D Vision Transformer, 3D Res U-Net, and 3D Attention U-Net-based deep models and their variants to provide diagnostic assistance for intracranial pathologies like aneurysms and ischemic strokes.
• Reduced the workload of radiologists and increased diagnostic accuracy for unruptured aneurysms by 10% with the assistance of developed models.
• Successfully developed and deployed Object Detection and Segmentation models for embedded devices such as UAVs, resulting in increased accuracy and efficiency in aerial surveillance and monitoring operations.
• Developed custom deployment pipelines and workflows that improved the deployment process of machine learning models, leading to a significant reduction in time-to-production.