Emily Carter

Operations analysis, from an unfinished source to a decision.

Information systems graduate, Texas A&M, class of 2026. Three analyses behind me. One of them is running in four clinics.

Role
Data or operations analyst
Where
Austin, Dallas, or remote
From
January 2027
Industry
Healthcare or logistics
Visa
US citizen, no sponsorship
At the capstone showcase

The work I want is the kind where somebody changes a schedule, a route or a reminder because of what the numbers said. All three projects below ended that way.

What I can do

Work with sources that are not ready

A live GPS feed with no documentation, a scheduling export with three date formats, a startup's database nobody had cleaned. Every project below began with data in that state.

See the transit project

Separate the cause from the first correlation

The clinic no-show work took three passes before reminder timing turned out to matter more than patient age, distance or insurance type, which were the expected answers.

See the clinic project

Present it to people who are not analysts

A transit office, a clinic operations manager, and two founders. In each case the useful output was a recommendation they could act on that week, rather than a model.

See the churn project

The work

Internship · Brazos Valley Health Partners · Summer 2025

Why patients were missing appointments, and what fixed it

The question

Two clinics were running a 14.7% no-show rate and losing about $22,000 a month in unused slots. The standing theory was that younger patients and longer drives explained it.

What I did

Cleaned 18 months of scheduling data, 22,400 appointments, into one table with reminder history attached. Tested age, distance, insurance type, appointment type, day of week, lead time and reminder timing. Age and distance explained almost nothing. Lead time and reminder timing explained a lot: appointments booked more than 21 days out with a single reminder were missed nearly three times as often.

What changed

I recommended two reminders, 48 hours and 4 hours before, for any appointment booked more than two weeks ahead. Operations piloted it in two clinics that September.

No-show rate by booking lead time 22,400 appointments, 18 months
10% 20% 9.1 8.4 11.4 9.0 16.8 10.1 24.6 11.7 Under 7 days 7 to 14 14 to 21 21+ days Booking lead time

One reminder Two reminders

What I would do differently: I spent my first two weeks building a dashboard nobody had asked for. The analysis that mattered was a single notebook and one chart.

Independent project · 2025

The campus bus was not as on-time as the schedule claimed

The question

The published on-time rate for the campus routes was 88%. Everyone who rode Route 12 in the morning doubted it, so I checked.

What I did

Pulled the public vehicle location feed every 20 seconds for eleven weeks, 1.2 million pings, and reconstructed arrivals per stop. The published figure counted a bus as on time if it arrived within five minutes of schedule at the route's first stop, which is the one place a bus can always be punctual.

What changed

Measured at every stop, Route 12 came in at 61% during the morning peak. I took the method to the transit office before I had a conclusion, rather than publishing it. They added schedule padding in January and now report the all-stop figure.

Screenshot, arrival map by stop

Contract work · Reveille Delivery · 2026

Finding the 200 customers worth winning back

The question

A campus meal delivery startup with 4,300 users wanted to spend a small credit budget on lapsed customers, and had no idea which ones to spend it on.

What I did

Built order histories into features that made sense for the business: days since last order, order gap trend, late deliveries experienced, and whether their usual restaurant had left the platform. I used logistic regression rather than a stronger model, because the founders needed to see why a customer had been flagged. A more accurate model they could not interpret would not have been used.

What changed

They sent credits to the 200 highest-scoring lapsed users and a random 200 as a control.

Walkthrough

Five minutes on the transit analysis

Where the feed came from, how arrivals were reconstructed, and the part where the first version of the method was wrong. Worth watching if you want to know how I think rather than what I concluded.

Video, 5 min 02 sec

We gave Emily an intern project and she came back with something we put into production. What stood out was that she told us which of her findings were weak. I have worked with analysts twice her age who do not do that.

Dana Whitfield, Operations Manager, Brazos Valley Health Partners

Tools, and how far I am with each

  • SQL Daily. Window functions, CTEs, query tuning on tables in the millions.
  • Python pandas and scikit-learn. Comfortable writing analysis; not a software engineer.
  • Tableau, Power BI Built and shipped dashboards people use. Tableau is the stronger of the two.
  • Statistics Regression, hypothesis testing, experiment design, including when a result is too weak to act on.
  • dbt, Airflow Used in coursework only. I can read a project, not run one in production yet.
  • Spanish Conversational. Enough to run an intake conversation, not enough for a clinical one.

Education and experience

  1. May 2026

    BS Information Systems, Texas A&M University

    Minor in statistics. Capstone: the transit reliability project above.

  2. Summer 2025

    Analytics Intern, Brazos Valley Health Partners

    Scheduling and capacity analysis across six clinics.

  3. 2023 – 2026

    Shift Lead, campus dining

    Thirty hours a week through school, scheduling eleven student staff. This is where the interest in operations started.

If you are hiring an analyst in January

I am happy to walk through any of these projects, including the parts that did not work. The fastest way to find out whether I am useful to you is to show me a question you are stuck on.

emily@emilycarter.com Resume, PDF

Page last updated 14 September 2026

Demo site by Resume Insite →