Dual B.S. in Astrophysics & CS
Rutgers University, 3.7 GPA. Society of Physics member, minor in Mathematics. The kind of person who thinks statistical mechanics was just ML before it was cool.
Data Scientist at Verizon · Open to opportunities
Astrophysicist. Data Scientist. Storyteller.
I used to study black holes. Now I find them in datasets instead. Both are equally mysterious, and both need a concerning amount of coffee.
Machine LearningXAI FrameworksCausal InferencePython & SQLGeorgia Tech MS
- $1.2B+
- Bad-debt impact
- 73%
- Faster hypotheses
- 3+ yrs
- At Verizon
- 2×
- STEM degrees
About The person
The person
behind the models
I'm Geet, a Data Scientist at Verizon who started out studying black holes at Rutgers and somehow ended up finding anomalies in billion-dollar datasets instead. Turns out the universe is full of patterns whether you're looking at galaxies or customer churn.
I'm currently a few semesters into my Georgia Tech Master's in Computational Analytics,
hanging in there by a thread token, but still going.
Outside of work: unhealthy amounts of anime, losing LP in League of Legends, and occasionally remembering to touch grass.
PythonSQLTensorFlowGCP / BigQueryStreamlitDataRobotApache SparkDockerTableauJava
Hosted the Verizon Analytics Summit
Presented at the 2023 summit alongside executives from Verizon, Bayer, BNY Mellon, NY Life, and ThoughtSpot. No pressure.
Trained ML on mice
At the Human Genetics Institute of NJ, built DeepLabCut models for kinematic pose estimation on wet-lab video. Yes, really.
Georgia Tech MS in Computational Analytics
My honest stance on the degree: check back in 3 years to see if I actually get that 4.0.
Experience Where I've been,
Where I've been,
what I've broken built
3+ years turning data into decisions at scale. Every number below is real.
Verizon
Nov 2024 to Present
Data Scientist III · Engineer III
- Architected a credit-risk fullstack app that synthetically generates a customer database to monitor policy changes impacting $1.2B+ in bad-debt writeoff.
- Engineered 9 Python & SQL anomaly-detection microservices with Verizon EDW integration, saving 50+ hours/month across 15 teams.
- Led 2 ML XAI frameworks (Streamlit + Qlik), reducing hypothesis-generation time by 73% for shareholder calls.
- Built churn-forecasting models and intervention simulators (SMS, email, fee adjustments) that pushed decisioning coverage from 0% → 60%.
- Presented to 100+ stakeholders, directly influencing product roadmap and reducing future development costs.
Verizon
Jan 2023 to Nov 2024
Data Scientist I
- Established a centralized DS pipeline framework/API for Corporate Finance on GitLab, Domino, and GCP, making collaboration 63% more efficient.
- Built a real-time AT&T and T-Mobile port-in EDA pipeline using DataRobot to forecast competitor market-share trajectories.
- Automated serverless forecasting for 9 major value brands (Walmart, Tracfone, Straight Talk, FiOS), pioneering a 12h email-report system.
- Developed Tableau dashboards on GCP/BigQuery, eliminating 10+ hours of weekly manual reporting.
- Ran K-Means customer segmentation that surfaced 3 at-risk cohorts, which directly informed new retention campaigns.
Verizon
Jun 2022 to Aug 2022
Predictive & Prescriptive Analytics Intern
- My first real exposure to Verizon's data infrastructure, and the internship that turned into 3+ years.
Human Genetics Institute of NJ
Oct 2021 to Jun 2022
Data Science Intern
- Trained DeepLabCut ML models for kinematic pose estimation on wet-lab mouse video, using Python, TensorFlow, and NVIDIA CUDA.
- Statistical validation via ANOVA and T-tests, benchmarked against published findings (Jones et al. 2020).
- Sponsored 4 Oxford iMARS Python extensions for retinal-cord tissue post-microscopy analysis.
RankSense
Jan 2021 to Oct 2021
Data Science Intern
- Researched SEO optimization across 100+ client domains, running A/B tests with Bayesian methods and CausalImpact.
- Improved click-through and baseline-conversion rates by ~32%.
Projects & Case Studies More projects
Things I've shipped
A mix of work projects (sanitized), academic deep cuts, and personal experiments.
01 · Live demo Personal
Credit-Default Scorecard (Live Demo)
A loan-level credit-risk model you can actually play with. Enter an applicant and it scores their probability of default right in your browser, then shows exactly which factors drove the decision. Built with weight-of-evidence binning and logistic regression (how banks really build scorecards), benchmarked against XGBoost with SHAP, and trained on public LendingClub data.
Try it live interactive, in-browser
02 · Live demo Personal
Uplift Modeling: Who to Persuade (Live Demo)
Predicting who converts is the easy question. This causal model answers the useful one: who converts only because you reached them. It estimates the incremental effect of a marketing email per customer using T-learner and X-learner meta-learners, validated with the Qini curve, then sorts everyone into Persuadables, Sure Things, Lost Causes, and Sleeping Dogs. Score a customer live, in your browser.
Try it live interactive, in-browser
03 · Live demo Personal
Significance: A/B Test Designer (Live Demo)
The experimentation toolkit behind a trustworthy A/B test. Power an experiment and size it before you launch, then analyze the results with a proper two-proportion z-test and confidence interval. Plus live demos of the two traps that quietly ruin most experiments: peeking (which inflates false positives past 25%) and skipping CUPED variance reduction. All computed in your browser.
Try it live interactive, in-browser
04 Work
Credit Risk Intelligence Platform
Fullstack app that synthetically generates customer databases to monitor credit-policy changes and their downstream impact on bad-debt writeoff. Built end-to-end (data pipeline, ML models, and a Streamlit frontend) so finance teams can simulate policy scenarios without ever touching production data.
$1.2B+ in bad debt monitored
05 Work
XAI Gross-Adds Framework
Explainable-AI framework isolating the real-time drivers of Verizon Gross Adds. Cut hypothesis-generation time by 73% for shareholder calls, so no more "we think it might be..." moments.
73% faster decision-making
06 Academic
Gravitational Microlensing Sim
Academic project simulating gravitational microlensing events, the same physics that bends light around black holes. Built in Python with Jupyter. The astrophysics degree, paying dividends.
github.com/geetpurohit
07 Work
Churn Prediction & Intervention Engine
ML models forecasting customer churn and simulating responses to interventions (SMS, email, fee adjustments). Pushed internal decisioning coverage from 0% → 60%, cutting outsourced engagement volume by ~35%.
0 → 60% decisioning coverage
08 Academic
DeepLabCut Pose Estimation
Trained computer-vision models to track mouse joint positions in wet-lab video experiments. Kinematic data benchmarked against published neuroscience research. The weirdest and most interesting project I've worked on.
Neuroscience × Deep Learning
09 Personal
AniParty: Video Sync Extension
Personal project: a browser extension that syncs video playback between computers. Built in JavaScript, because watching anime with friends across the country shouldn't require a scheduling meeting.
Solving the important problems
Skills
The toolkit
Proficiency built by shipping things, not just finishing courses.
Core Strong Working knowledge distance from the core = depth
Languages
- Python
- SQL
- R
- Scala
- Java
ML & Modeling
- scikit-learn
- XGBoost / LightGBM
- PyTorch
- TensorFlow
- Hugging Face / LLMs
Data & Cloud
- BigQuery / GCP
- Apache Spark
- Airflow
- dbt
- Docker / Kubernetes
Experimentation
- A/B Testing
- Statistics
- Causal Inference
- Looker / Tableau
- MLflow
Contact Let's talk.
Let's talk.
Seriously.
Whether it's a job opportunity, a collaboration, or you just want to argue about whether anime is peak storytelling, my inbox is open.
[email protected]