About the role
A delivery forecast can look excellent if tomorrow’s corrections slip into yesterday’s dataset. Working with an experienced engineer, you’ll prepare time-stamped observations and test what a model could realistically have predicted.
You’ll begin with short-horizon demand forecasts, where a planner often needs tomorrow’s estimate before today’s reports have finished arriving. You’ll help the team test which information it can safely use at that deadline and explain the resulting forecast in the units a planner works with.
Working at ENNPRO
ENNPRO is an applied machine learning consultancy. We work alongside client engineers and the people who use their systems. Depending on the engagement, a client might need a service they can operate, a benchmark they can reproduce or enough evidence to decide whether to release a model. Their team should be able to use the result after handover.
This is a hands-on role for someone learning how forecasts are used outside a notebook. You’ll work with an experienced engineer who can review a split, talk through an error and help you decide what to try next. You’ll also hear directly from the person who uses the forecast. They may care about the number of staff needed for a shift, a minimum stock level or how much warning they get before a busy day.
You’ll own pieces of an experiment that another person can check. Early on that might mean preparing a clean daily series and a baseline. As you become familiar with the data, you’ll take a defined question from the original records through to a result you can present to the team. Questions about confusing timestamps or an unexpected zero are part of the work.
Responsibilities
- Prepare dated observations in Python and check a sample against the source. Flag missing dates, revised values and measurements that arrived after the prediction cut-off.
- Implement a simple daily or seasonal baseline. Write a small test by hand so that a teammate can see exactly which earlier observations are allowed into a forecast.
- Run rolling backtests at the lead time the planner needs. Report errors in familiar units and plot a few of the dates where the model misses badly.
- Work through a code review with a teammate. Turn a surprising input or result into a regression test, then explain the change to someone who uses the forecast.
- Document the unit of each series and what a single row represents. Check whether a zero means that nothing happened, the site was closed, or the feed had not arrived when the record was created.
- Prepare a chart and a short set of examples for a planning review. Label the forecast date, its lead time, the observed outcome and the size of the miss so a planner can interpret it without reading your code.
- Keep experiments reproducible. Save the configuration, baseline definition and data dates alongside the result, and give a teammate enough information to rerun the same comparison from a clean checkout.
Day to day
Most of your work will use Python, tabular data, simple plots and version control. Start with a short period that you can check manually. Calculate a small forecast with a pencil or spreadsheet and then confirm that the program gives the same answer. When it differs, show your reviewer the first row where the two calculations diverge.
You’ll learn to keep a clear log of the questions that came up while preparing the data. One good question for a source owner can be more valuable than several rounds of tuning. If a key field changes at midnight in a different time zone, finding that detail belongs in the project notes and the tests.
Projects and first 90 days
A representative assignment
A shift planner asks for tomorrow morning’s demand. Start with yesterday’s count and the same weekday last week. On a calendar, mark when each input became available. Compare the two rules on the same sequence of forecast dates.
Bring one particularly bad day to review. An experienced engineer will help you establish whether it reflects a genuine demand change, a late reading or an error in the replay. Your notebook and test should make that distinction inspectable.
The planner might point out that the forecast for a holiday was technically close but arrived after the staffing decision. Add the decision deadline to the replay and check how many forecasts were usable on time. That is a practical way to connect model evaluation to the work it is meant to support.
How your first engagement could unfold
First month. Pick a small historical series with your reviewer. Learn how to set up the project, read the source records and run an existing test. Reproduce the daily totals for a short period and write down the questions you would ask the person who collected them.
Second month. Build a baseline and a rolling evaluation with a teammate reviewing the date boundaries. Take a few large errors to the planning team and listen to how they explain the days involved. Use one of those explanations to improve the dataset notes or add a missing test.
Third month. Own a small comparison from beginning to end. Present the result in the units the planner uses, explain a limitation and identify one sensible follow-up. Hand the code to a teammate and make the changes they need in order to run it themselves.
Qualifications
Core experience
- You can clean and join a small dataset in Python and explain one modeling project end to end. Coursework, independent work and professional projects are all relevant.
- You understand why shuffling dates can give a misleading forecast score and can explain a baseline in ordinary language.
- You’re comfortable sharing unfinished work, asking for a review and documenting a mistake so the next person can catch it earlier.
- You can use a plot or a small table to investigate an unexpected result and describe your reasoning step by step.
- You can break a piece of work into a manageable first version, take feedback on your code and explain what you changed. Experience doing this in a class or personal project counts.
Useful additional experience
- Familiarity with pandas, NumPy, SQL or a similar tool for organizing dated observations.
- Experience with a seasonal baseline, a project involving missing observations, or a comparison that used several historical cut-offs.
- Practice sharing a notebook, a pull request or a short written analysis with someone who needed to understand the result.
How to apply
Walk us through a forecast or data project you built. Describe one error you investigated. A short explanation is enough if you can’t share the underlying data.
A modest project is fine. Tell us what the data looked like, what you predicted and how you checked whether you were right. If a course or team supplied part of the solution, tell us which part you built. We’re particularly interested in the moment you noticed something that deserved a closer look.
You can use an anonymized example for confidential work. The application form includes space for your account of the project, an optional résumé and an optional link to a work sample. It currently lets you prepare and review a draft in your browser.
Equal opportunity and accessibility
ENNPRO welcomes applicants of all backgrounds. We consider candidates on the skills and experience relevant to the role, including experience gained through different educational and career paths. If you need an accessible version of these materials or an adjustment to the application process, write to contact@ennpro.pro.
Start application