Case Study: Student Outcome Intelligence Platform 
Context 
You are joining as data scientist/engineer at a large Norwegian university with 35,000 
students across 8 faculties.  
The university generate data across several distinct operational system. To give you 
enough context to prepare - without prescribing your solution - here is the description of 
what those systems typically contain:  
Student Information System 
The core administration record - enrolment status, programme, faculty, grades, personal 
details, graduation and withdraw history. The university system of records for who a 
student is.  
Learning Management System 
The digital learning platform - login activity, course access, assignment submission, 
discussion forum participation, video watch time, quiz attempt. High-frequency 
behavioural data.  
ERP System 
Financial aid records, tuition payments status, scholarship awards, outstanding 
balances. Separate from the student system - often a different vendor entirely. 
Physical Campus System 
Library access logs, WiFi presence data, building entry events.  A signal of physical 
campus engagement independent of academic records.  
These systems do not talk to each other. Data formats, update frequencies and identifier 
different across all of them. Some export file, some have API, some emit real time 
events.  
The Goal 
The university want to identify students at risk of dropping out before it becomes visible 
in any official record - early enough for an academic advisor to intervene.  
Right now, it is not possible. The signals exist across separate systems, but nobody see 
all of them together. A student can be missing lectures, disengaging online, and falling 
behind financially - and each system see only its own fragment. By the time a grade 
report confirms the problem, the window to act has usually closed.

Your task is to design and build the platform that closes this gap; get the data in from 
four systems, make it trustworthy, use it to predict risk and deliver those predictions to 
advisors each semester in time to act.  
The platform handles personal data about real students. It needs to be accurate, 
explainable, fair, and auditable. It also needs to keep working correctly long after launch.  
What to factor in 
* The four systems were built independently, for different purposes, over many 
years. Their data is inconsistent and their schema change without notice. 
* Students change status over time - they take leave, switch programme, withdraw 
and return. A platform that only holds current state will build incorrect history 
and product unreliable predictions. 
* Predictions must be made using only information that existed at the time of 
prediction. Using future data during training produces a model that cannot work 
in production.  
* The student population is diverse. Engagement patterns differ across groups for 
reasons unrelated to dropout risk. A model that does not account for this will 
systematically flag the wrong students.  
* At some point a student or regulator will ask why that student was flagged red 
and who has access to their data. That question must be answerablr from what 
you built.  
Your task 
Prepare a 20-30 minute presentation where you describe how you would design and 
build this platform. 
We assume you will spend approximately 2-3 hours preparing your response. 
Focus on the key decisions needed to make the solution work in practice, rather than 
covering every detail. We are not looking for a single "correct" solution, but for how you 
approach the problem and the choices you make along the way. 
You may use diagrams where helpful. No coding is required. 
Be explicit about: 
* assumptions you make  
* trade-offs in your design