Vasyl Shafran

Designed and deployed Generative AI solutions by integrating language models such as OpenAI GPT-4 into interactive applications, enabling automated content generation, summarization, and conversational capabilities. Fine-tuned domain-specific LLMs to improve relevance and accuracy for enterprise use cases, achieving a 17% increase in model performance. Architected retrieval-augmented generation (RAG) pipelines using vector databases (Pinecone, Weaviate) and embedding models (OpenAI, SentenceTransformers) to deliver accurate, context-aware responses in real time. Developed and fine-tuned advanced LLMs using PyTorch, improving model accuracy by 15% through optimized training techniques and hyperparameter tuning. Led a project analyzing patient feedback with text-processing approaches to improve service delivery in healthcare and leveraged ML on patient records to inform personalized treatment plans.

Designed and deployed Generative AI solutions to enhance customer experience by integrating language models such as OpenAI GPT-4 into interactive applications, enabling automated content generation, summarization, and conversational capabilities. Fine-tuned pre-trained LLMs on domain-specific data to improve relevance and accuracy for enterprise use cases, achieving 17% increase in model performance. Architected RAG pipelines using vector databases (Pinecone, Weaviate) and embedding models (OpenAI, SentenceTransformer) for real-time, context-aware responses. Developed and fine-tuned advanced LLMs with PyTorch, boosting model accuracy by 15% through optimized training and hyperparameter tuning. Led a real-time text processing project to analyze patient feedback in healthcare, improving service delivery, and leveraged ML techniques to identify patterns in patient records to inform personalized treatment plans.

Deployed sentiment analysis models using NLTK on customer feedback data with 94% accuracy. Developed MySQL triggers for real-time data validation, reducing data entry errors by 56% and maintaining regulatory compliance. Created data analysis pipelines using Pandas, enabling the team to process 4x more data. Executed keyword extraction and document classification with unsupervised LDA. Conducted data analysis and model training using AWS SageMaker, integrating Python and AWS Lambda for automated, scalable solutions across multiple business functions. Engineered a robust CI/CD pipeline using Google Cloud Build, which reduced time-to-market for new features by 50%.