My Projects
These are applied projects that start with a practical question and end with something a
decision-maker could use. They span healthcare access, actuarial, mobility, operations, iGaming,
and sports forecasting, and they draw on
Python, R, SQL, Tableau, Looker, Power BI, and GCP, along with statistical
modelling and machine learning. Projects link to source code, reports, dashboards, and case
studies where available.
Mobility · Time series · Python · GCP · BigQuery
EV battery performance analysis
Electric vehicles are a cornerstone of cleaner transport globally. In East Africa, electric
motorcycles (boda bodas) are especially promising as public transport, carrying passengers
across both busy urban centres and rural areas where other options are limited. Their climate
benefit, however, depends on how sustainable they are, i.e. how long the batteries last
and how well fleets are run.
I analysed battery swap and telemetry data to understand how batteries degrade with use, then
built time series forecasts of fleet usage.
findings on degradation patterns and usage trends show how operators could plan swap
capacity, maintenance, and replacement more effectively. Longer battery life means fewer
batteries manufactured and discarded, and better-planned swapping means less idle
infrastructure, which makes electric mobility more economical and more sustainable.
Healthcare · Python
Mapping healthcare access gaps
Access to care varies widely across Nairobi, and averages hide where the shortfalls are.
This project compares sub-counties on facility access, bed availability, and selected health
services, measured against population size and recommended target ratios.
The analysis highlights which areas are underserved and by how much
several sub-counties fall below the UN Department of Economic and Social Affairs
SDG-recommended targets,
giving planners and health organisations an evidence base for prioritising resources.
Operations · SQL · Python · Excel
Team Performance & Insights
I analysed data from roughly 170,000 cases to assess an operations
team’s performance against its “instant service” promise. Using SQL,
I cleaned inconsistent timestamps, then measured backlog age, time to first review,
resolution time, reopen rates and agent throughput, and presented the results as a deck.
Cases entered the backlog within 45 seconds and reviews took about
11 minutes, yet cases waited 18 hours for a first review,
driving a 72-hour average resolution time. The reopen rate was
31%, and resolution time worsened over time even as the resolution rate
improved.
My analysis identified six cases with 200+ hour backlog delays and
recommended fixing the first-review delay to improve resolution time and
reduce reopen rates.
iGaming · Tableau
iGaming Dashboard
I built an interactive Tableau dashboard for an iGaming operator, tracking
active players, deposits and net gaming revenue (NGR) from February to April,
alongside the registration-to-deposit funnel, revenue by product and performance by market.
Active players grew from 3,663 to 4,092 and deposits rose to
€2.27M, but NGR fell in April to €586K.
Casino generated about 75% of revenue, and
Sweden, Finland and Norway led the markets. Bonus spend was positively
correlated with NGR, turnover and active players. I also flagged
Malta and Hungary as loss-making markets with growth potential.
Mobility · Looker · GCP
Ride-Hailing Dashboard
I built an interactive Looker Studio dashboard analysing ride-hailing
supply against demand for the week of 7–13 November 2016. It tracks
searches, zero-car searches, coverage ratio and driver hours, with an hourly coverage
heatmap and filters for day, hour and coverage level.
Of 7,047 searches, 1,858 (26.4%) found no car, giving
an average coverage of 74%. Coverage was weakest in the
early hours and dipped on 11 November, when searches
peaked. Only 24% of online hours (21,288 of 88,656) included a booking,
which suggests a
mismatch between when and where drivers are online and when demand occurs,
rather than a simple shortage of drivers.
Actuarial · Python
Claims reserving with chain ladder
Insurers need to estimate what claims will ultimately cost before they are fully settled.
Using payment data, I built a claims development triangle, calculated loss development
factors, and projected ultimate losses with the chain ladder method. The project walks
through each step of the reserving process, from data preparation to the final reserve
estimate, in clear, reproducible code.
Sports · ML · Python · SQL
Football match outcome predictor
A web app that estimates match outcomes across leagues and teams. Instead of treating all
history equally, the model uses recency-weighted performance, so recent form counts for
more than results from seasons ago. It shows an end-to-end workflow from data and modelling
through to a usable interface.
Actuarial · R
Claims reserving in R
A companion to the Python reserving project, implementing the same chain ladder approach
in R. Having both versions allows a direct comparison of the two workflows and shows the
method working in either language, which is useful for teams whose actuarial work is split
between them.