About Me
I am a passionate data scientist with a strong foundation in machine learning, NLP, and data analytics. With experience spanning academia, healthcare, and government projects, I thrive on solving complex problems with innovative solutions. I love learning, adapting to new challenges, and leveraging data to drive meaningful impact. Whether it's fine-tuning AI models, automating workflows, or visualizing insights, I bring curiosity, creativity, and a problem-solving mindset to everything I do.
Skills
What I bring to the table
Education
University of Michigan
Ann Arbor, MI, USA
Master's in Data Science
Achievements: GSI with tuition waiver (SI649), research assistantships, Ross Hackathon (2024 & 2025), TAMU Healthcare Hackathon
- Probability and Distribution
- Statistical Inference I & II
- Machine Learning, Regression, Time Series
- Causal Inference, Data Analytics & Viz
Savitribai Phule Pune University
Pune, MH, India
B.Tech in Information Technology
Achievements: Event Head at Codechef, TedxVIIT, Finance/Sponsorship, Sports Club Rep
- Machine Learning, AI, Discrete Maths
- OS, Networking, Data Structures
- C/C++, Java, HTML/CSS, JS, PHP
- Object-Oriented Programming
Experience
Data Scientist – Michigan Medicine
Ann Arbor, MI
- Implementing the DECI model (Microsoft Causica) to generate and analyze causal graphs and compare produced adjacency matrices with those generated using the DoWhy library to evaluate model accuracy and consistency
- Assessing the reliability of synthetic data by analyzing and comparing adjacency matrices of original and synthetic datasets through quantitative techniques such as network centrality and graph similarity
- Developed a semantic search pipeline using Retrieval-Augmented Generation (RAG) on ASO guideline documents, leveraging LangChain and vector stores. Experimented with multiple text splitting strategies to optimize chunking for improved retrieval accuracy in domain-specific LLM tasks.
Data Analyst – UM Precision Health
Ann Arbor, MI
- Increased survey participation rates by 50% by curating surveys and developing SQL queries and Python-based analytics to process and extract insights from response data, identifying key behavioral patterns and trends
- Developed real-time Power BI dashboards for survey monitoring
- Implemented reinforcement learning-based interactive games to assess patient attention span, collecting structured behavioral data for deeper analysis
Research Assistant – School of Information, UMich
Ann Arbor, MI
- Achieved 98% word retrieval accuracy by developing a web scraping model for structured data extraction from American Historical textbooks
- Increased classification accuracy by 10% by expanding spaCy NER annotations with 7 new historical categories
- Cut manual validation time by 2 hours per dataset by creating an algorithm for data annotation verification
- Pre-processed and trained LLMs model on OntoNotes 5 and a custom data, fine-tuning GPT-NER and Meta-Llama-3-8B-Instruct using LoRA and 4-bit quantization, achieving 66% NER accuracy, improving historical NER
Graduate Student Instructor – SI649
Ann Arbor, MI
- Taught data visualization using Tableau, Plotly, Altair, D3.js, and GenAI tools.
- Led interactive design labs with real-world datasets.
Data Science Intern – Ministry of Defence
Pune, India
- Enabled real-time skeletal tracking for army drills by developing an AI-based pose estimation model using OpenCV, Mediapipe, and YOLOv8, tailored for military movement analysis
- Deployed a real-time feedback system generating automated pose correction reports and annotated video snippets, assisting trainees with immediate visual guidance. Deployed the algorithm using Flask
- Designed and Created UI/UX using Figma and user-friendly web application using Flask. Paper submitted to IEEE
Data Scientist – Beyond Business Travel
TUS, Ireland (Remote)
- Built a data extraction pipeline for Enterprise and Hertz invoices using OCR, NLP, and machine learning, achieving 98% accuracy and reduced manual entry time
- Outperformed RPA tools such as BluePrism and UIpath with improved anomaly detection and 75% faster processing.
Projects
🧠 VAE Variants for Causal Effect Estimation
Nov 2024 – Dec 2024
- Extended the CEVAE model to improve causal inference under unmeasured confounders using PyTorch and TensorFlow
- Implemented VAE variations, including Correlated-V AE, Beta-VAE, HVAE, and VQ-VAE, for enhanced latent variable modeling
- Conducted experiments on real-world (IHDP, JOBS, TWINS) and synthetic datasets using Pyro and Adamax optimizer achieving the best PEHE of 1.47 ± 0.18 and ATE error of 1.26 ± 0.75 on the JOBS dataset
⚽ Soccer Analytics Dashboard
Apr 2024
- Analyzed and processed over 3 million soccer records from 11+ datasets using SQL and Python, performing data cleaning, feature engineering, and statistical evaluation
- Developed an interactive Tableau dashboard to visualize key soccer analytics
🦠 TB Forecasting with ARIMA and POMP
Nov 2023 – Dec 2023
- Improved TB incidence forecasting accuracy by comparing ARIMA and SEIRS-based POMP models
- Implemented a stochastic POMP model with overdispersion and time-varying transmission, achieving a more realistic fit to TB decline trends over time
🦈 Shark Tank Deal Prediction
Oct 2022 – Dec 2022
- Identified key trends in startup features (industry, revenue, valuation) by performing exploratory data analysis using Python (Pandas, Matplotlib, Seaborn) and R
- Achieved an F1-score of 87.09% by building a machine learning pipeline with ANN, outperforming SVM and Random Forest in predicting startup funding offers
- Improved post-investment evaluation by implementing a fuzzy logic system in MATLAB with 22 rule sets to categorize deal quality
Publications
Offer and Deal-Quality Prediction using Machine learning and Fuzzy approach: A Shark Tank India Case Study
Proceedings of the ACM Web Conference 2023
Link 2023 Published
Empowering India’s Climate Action: Harnessing Blockchain for Carbon Trading
2024 IEEE International Conference on Blockchain and Distributed Systems Security (ICBDS)
Link 2024 Published
Contact Me
(shreyadj@umich.edu)