Highly accomplished PhD candidate with over 5 years of experience in Data Science, specializing in AI, Machine Learning, and Natural Language Processing. Proven ability to build and deploy complex ML solutions using Python, TensorFlow, and Hugging Face, consistently delivering production-ready tools and leading impactful research initiatives.
Remote, N/A, US
→
Summary
Currently developing and optimizing AI-powered conversational agents for healthcare applications, focusing on robust data retrieval and model refinement.
Highlights
Built a Retrieval-Augmented Generation (RAG)-based chatbot for Wound Tele.AI Pro, leveraging LlamaIndex, ChromaDB, and Hugging Face to provide evidence-backed answers to wound care queries.
Developed a semantic chunking and preprocessing pipeline for dense passage retrieval across PubMed articles, enhancing data preparation for AI models.
Evaluated 5+ embedding models and chunking strategies on 50+ wound-related queries, improving precision@5 by 17%.
Prototyped a local QA tool using Haystack, ChromaDB, Hugging Face, and Streamlit for efficient offline testing and user validation.
Collaborated with wound care experts to validate chatbot output and refine domain alignment for patient use cases, demonstrating effective leadership and cross-functional teamwork.
Madison, WI, US
→
Summary
Led an 8-member research team in astrophysics, applying advanced data science and statistical modeling to analyze large-scale stellar datasets and generate significant scientific insights.
Highlights
Led an 8-member team to author a peer-reviewed study on vertical kinematics of the Milky Way, utilizing Gaia survey data.
Queried, cleaned, and integrated 500,000+ stellar records using SQL and Python; applied bootstrapping, correlation analysis, and hypothesis testing to extract vertical motion patterns.
Validated findings with greater than 3-sigma confidence, demonstrating end-to-end execution from data engineering to insight generation.
Mumbai, Maharashtra, India
→
Summary
Engineered features and applied machine learning models to classify particles from a large-scale simulated detector dataset, achieving high accuracy.
Highlights
Engineered features from a 500k-instance simulated Belle detector dataset, applying a Boosted Decision Tree model to classify low-momentum muons from background particles.
Achieved 85% accuracy in classifying low-momentum muons, significantly improving particle identification.
Python, SQL.
Pandas, NumPy, Scikit-learn, TensorFlow, Hugging Face Transformers, Streamlit, Power BI (basic), Git, LlamaIndex, Haystack, ChromaDB.
Deep Learning, NLP, Retrieval-Augmented Generation (RAG), LLM Fine-Tuning, Embeddings, Vector Search.
Docker, Google Cloud, Snowflake, dbt, Airbyte, GitHub.
Data Wrangling, Data Cleaning, Feature Engineering, ETL Pipelines, Statistical Modeling.
→
Summary
Developed and deployed an auditing and interpretability tool for gastric cancer histopathology, leveraging deep learning to classify gastric cancer tumor microenvironment tissue images and enhance diagnostic trust.
