Engelhart Software Engineer Interview Experience — Quant Developer Second Round, Data Processing and Weather Forecast System Design

I recently interviewed for a Quant Developer role at Engelhart, on the US Power & Gas side.

The second round was two hours, split into data processing and system design.

The data processing part was done in something like Colab, but it was super annoying: the file the interviewer wanted me to open wouldn't load (404), so I had to quickly download some time series + node data from a website, hourly level. The main problem was that I'm not familiar with pandas. The LeetCode-style stuff isn't enough. You need to be comfortable with real-data work, especially finding missing values in a multiindex and data transformation. As a result I basically didn't finish anything.

The system design was done on Miro. The interviewer said clearly that no code was needed, I mainly just had to draw the architecture / flowchart and add some pseudocode.

The gist of the question: around 6 AM every day, a public URL publishes a weather forecast CSV. The data includes:

• 10 weather scenarios
• the next 15 days
• hourly frequency, so about 360 timestamps per scenario
• weather variables such as temperature, wind, soil moisture, cloud/rainfall

The system already has a set of black-box models provided by a quant research team, for example:

Weather inputs
├── Electricity Demand Model
├── Wind Power Model
├── Solar Power Model
└── Hydro Power Model
        ↓
Electricity Price Model

I was asked to design a system that does: daily forecast ingestion → validation → historical storage → run every model across all scenarios → aggregate results → make the forecasts available to traders in a dashboard. They also wanted results generated as soon as possible after a new forecast is published.

After that the interviewer started drilling down layer by layer:

• What stack would you use for the ingestion service? How would you schedule it given that the file arrives around 6 AM rather than exactly at 6 AM?
• Would you store the CSV exactly as received or transform it?
• One DAG or multiple tasks?
• How do you kick off the model runs? Are these part of the same Airflow workflow?
• Where could this pipeline fail, and how would you handle different failures?
• Summary aggregation
• How would you materially make this available to traders? What dashboard stack would you use and how would you host it?
• Right now there are only 10 scenarios. If it later becomes hundreds or even thousands of scenarios, what would you change?