Work Experience

Data Scientist at Astrazeneca

September 2023 – Present

Gaithersburg, MD

2-years rotational research program within R&D department, focusing on Data Science & AI

Placement No. 1: Developed a PofC automated pre-screening pipeline for homologous recombination deficiency (HRD) in ovarian cancer patients, using biopsies imaging data, to improve timelines of disease diagnosis (Tech: Python (Pytorch, tensorflow, scikit-learn), High-performance computing, bash scripting, GitHub)

Key Responsibilities & Achievements

• Elevated HRD prediction accuracy among ovarian cancer patients by 23% PPV using pre-screening biopsies
• Developed a reusable python program for model training, validation, and testing for user-defined imaging data—including cross-validation, metrics computation, and visualization (ROC AUC, PR curves, and attention heatmaps)—using scikit-learn, matplotlib/seaborn, and PIL
• Led and organized team meetings with pathologists and data scientists to refine the design and the interpretability of the PofC AI-driven diagnostic tools
• Contributed to an open-source Python package (slideflow) by modifying the processing pipeline and implementing a customizable feature extractor class through Pytorch, used git version control to track progress
• Presented work at the 2024 AstraZeneca Data Science Symposium, 2024 Graduate Program Symposium, finished the manuscript and is working on publication approval, codes available for review on company GitHub

Placement No. 2: Developed a proof-of-concept chatbot utilizing agentic AI, RAG and LLMs to improve clinical trial efficiency (Tech: Python (LangChain/LangGraph), AWS Bedrock, AWS Sagemaker, AWS Sandbox, FastAPI, GitHub)

Key Responsibilities & Achievements

• Evaluated and benchmarked of State-of-the-Art LLMs on AWS Bedrock and Retrieval-Augmented Generation (RAG) components for unstructured clinical protocols (CSPs), informing product pipeline architecture to meet both technical and business requirements
• Architected and deployed a proof-of-concept, LLM-powered chatbot using AWS Bedrock agent (e.g. Claude 3.5 Sonnet, Titan’s Embedding), LangChain, and LangGraph, running via FAST API, enabling customized clinical trial protocol development enabling unstructured documents digitalization, objectives, endpoints, assessment table etc. extractions
• Developed and integrated LLM-based evaluation metrics (precision, recall, answer accuracy, hallucination) using Graph API via LangGraph to ensure robust model performance and reliability
• Designed and implemented a simple RAG system for automatic mapping of clinical assessment ontology, leveraging Text Embedding models, achieved greater than 70% precision@1
• Collaborated effectively in an Agile environment with a global team, utilizing Jira for sprint planning and issue tracking, ensuring efficient project execution and delivery
• Presented work at digital health department data science group meeting and 2024 Graduate Program Symposium

Placement No. 3: Drafted a study protocol and performed statistical analysis using real-world data (RWD) of claims records and EHR data to inform clinical trial designs in Early R&I (Tech: R (ggplot2), PostgreSQL, AWS Redshift)

Key Responsibilities & Achievements

• Independently authored a study protocol and statistical analysis plan (SAP) for a retrospective, observational real-world data (OPTUM claims data), investigating multimorbidity in patients with asthma, COPD and bronchiectasis
• Designed a data processing and analysis pipeline using SQL queries on Amazon RedShift and wrapped with reusable R functions, enabling accurate cohort definition, statistical and epidemiological analysis, tracked changes via Git
• Providing key insights using boxplots, Sankey plots etc., generated using ggplot2 visualization, provide important numbers for clinical trial decisions in early R&I for respiratory diseases
• Currently conducting a parallel analysis on TriNetX (EHR) data to inform study development across respiratory diseases and validate the harmonization across data, using the established study protocol and SAP as guide for analysis, ensuring methodological consistency

Part-time Data Analyst at Columbia University

October 2021 – February 2025

New York, NY

Performed end-to-end clinical data analysis, leveraging advanced machine learning and deep learning models to enhance risk assessment (Tech: R, Python (tensorflow))

• Conducted comprehensive data cleaning, feature selection, EDA and regression analysis on multi-center perioperative clinical data from the anesthesiology department using R
• Enhanced the predictive power of traditional risk assessment models by incorporating patient multimorbidity through the application of machine learning (Random Forest, XGBoost) and deep learning models
• Implemented Shapley value analysis in Python to provide model interpretability and prepared a manuscript incorporating anesthesiologist feedback
• Generated forest plot, violin plots etc. and publication-ready tables utilizing ggplot2 and gtsummary packages

Data Analyst and Machine Learning Summer Internship at Yrobot Inc.

May 2022 – September 2022

Boston, MA (Remote)

Developed robotics data analysis and visualization Python program (Tech: Python)

• Developed a custom Python data preprocessing and visualization tool using NumPy, Matplotlib and Plotly that standardized output and streamlined the analysis of IMU sensor data for patient walking patterns
• Collaborated with the engineering team to customize an open-source RNN model for 3D human motion estimation, using company data to propose and prioritize experimental directions

Bioinformatics Student Researcher at Columbia University

February 2025 – June 2025

New York, NY

Performed bioinformatics analysis using RNA-seq data (Techniques: Bash scripting, Python)

• Performed RNAseq Data analysis via BLAST for read trimming and alignment
• Performed pathogen identification on lab virus sample using bioinformatics tools VirCapSeq-VERT
• Help developed Nextflow usage documentation for lab members