Venkata Sai Ujwala Bayana

Master's

Data Science

BOEHRINGER INGELHEIM

Data Scientist

• Implemented a Streamlit application utilizing Azure Open AI GPT 4 and Databricks to generate a summarized Regulatory Document for all EU Medical Trails. Successfully presented the outcomes to BI’s European Union Regulatory Team, resulting in approval for production. • Optimized content significance by deploying various chunking strategies such as Content-aware (NLTK, spaCy, Markdown, Recursive) and Fixed-size Text splitters via LangChain leading to a 86% improvement in performance. • Enhanced semantic or context-based search by amalgamating Pinecone Vector Store with LlamaIndex and implementing Recursive Retrievers to explore ‘Top-K’ similar nodes, resulting in 47% faster contextual data retrieval. • Utilized Prompt engineering (Zero-shot, Few-shot, CoT) and question answering chains (load_qa_chain, RetrievalQA) to interact and instruct LLMs to extract contextual responses, thereby reducing ambiguity of response by 37%. • Accelerated drug discovery by 80% by engineering ML pipeline on AWS, Databricks leveraging antiBERTy, autoencoders and XGBoost to classify potent cancer-fighting antibodies, based on sequence and monomer data, with a F1-score of 0.87 to identify promising sequences. • Reduced sales estimates variability by 30% and downtime by 40% through a CI/CD pipeline on AWS automating forecasting process using hierarchical time series models, updating real-time QuickSight dashboards for finance and sales.