Data Scientist

Jim Halpert

Data Scientist focused on forecasting and classification problems that actually ship. I build models that predict things. Mostly they predict that Dwight will fall for it. He does.
Open to workParis, FRingh
5
Projects
8yrs
Experience
10
Skills
Expert 2Advanced 4Intermediate 4

Projects

01Featured project

A/B test analysis with statistics

Analysed an A/B test on a checkout page: sample size check, two-proportion z-test, confidence intervals, and a bootstrap to double-check the result. The variant looked like a 4% lift but the interval crossed zero; recommended running two more weeks instead of shipping. Also simulated the central limit theorem to show why the test needed that much traffic.

PythonNumPyStatistics
#ProjectSkills used
02Loan default classificationTrained and compared logistic regression, random forest and gradient boosting on a consumer-credit dataset with 5-fold cross-validation. Handled the class imbalance with class weights rather than oversampling, chose the threshold on the business cost of a missed default, and explained the final model with SHAP values.
2
03Weekly sales forecast for 40 storesDecomposed three years of weekly sales into trend, seasonality and holiday effects, then compared seasonal ARIMA, Prophet and a LightGBM model with lag features on a rolling backtest. The gradient-boosting model won on 31 of 40 stores; Prophet was more robust on the small ones. Delivered a notebook and a one-page summary.
3
04Product image classifier in PyTorchTrained a small CNN from scratch, then a transfer-learning model on ResNet-18, to classify product photos into 12 categories for a catalogue. Transfer learning reached 93% accuracy against 81% for the custom network, with a tenth of the training time. Tracked runs with MLflow.
2
05Bike-crash risk map of Paris · PersonalCombined the national road-accident open data with the Paris cycling network to estimate crash risk per road segment, normalised by cycling traffic counts. Built a small Streamlit app where you pick a route and see the riskier intersections. Mostly an excuse to learn geospatial joins.
3

Skills

BeginnerIntermediateAdvancedExpert
Projects
Python
Pandas
scikit-learn
Statistics
Time series
SQL
PyTorch
Streamlit
Airflow
MLflow

Experience & education

20152017201920222024Now
Data Scientist · BlaBlaCarWork2024 — Present

Demand forecasting for the bus network: weekly forecasts per route used by the pricing and capacity teams. Replaced a seasonal baseline with a gradient-boosting model, which reduced the forecast error by about 18% on the top 50 routes. Python, scikit-learn, Airflow, BigQuery.

Junior Data Scientist · EkimetricsWork2022 — 2024

Consulting on marketing-mix models for two retail clients: data preparation, model fitting, and the monthly results deck for the client's marketing director. Learned to explain a coefficient to someone who does not care about coefficients.

MSc Statistics and Machine Learning · Université Paris-SaclayEducation2021 - 2022

Thesis on probabilistic forecasting for intermittent demand.

Data Scientist path · D8A AcademyEducation2021

Four validated projects: statistics, supervised learning, time series, deep learning.

Sales Representative · Dunder MifflinWork2018 — 2021

Paper sales. Built a small forecasting spreadsheet for my own accounts, got curious, and went back to school for statistics.

BSc Mathematics · Université de LilleEducation2015 - 2018

Skill coverage

Pandas
2 / 5
Python
2 / 5
CNN
1 / 5
NumPy
1 / 5
Prophet
1 / 5
PyTorch
1 / 5
scikit-learn
1 / 5
Statistics
1 / 5
statsmodels
1 / 5
Streamlit
1 / 5
Time series
1 / 5
Number of projects using each skill, out of 5.