EliseAI Data Scientist Interview Experience — Tic-Tac-Toe Coding Round and a Lease Renewal Take-home

I applied online and a recruiter reached out.

The first round was a 30-minute chat with a team member.

The second round was coding, a tic-tac-toe variant. I'd suggest looking at all three of these:
348 - Design Tic-Tac-Toe
794 - Valid Tic-Tac-Toe State
1275 - Find Winner on a Tic Tac Toe Game

The third round was a take-home: you get 24 hours to do it, and then there is a 30-minute presentation. The task covers one supervised learning problem and one unsupervised learning problem.

Objective: Build a model predicting likelihood for lease renewal. Instructions:

Data: You will have access to two main forms of data: i. JSON files containing 1000 conversations of tenants considering renewing their leases; ii. A SQLite database consisting of three types of tables accessible at this link. The tables are summarized as follows: residents: resident information; maintenance_history: maintenance work order submissions by resident; renewal_offers: renewal offers provided to each unit. Both i and ii are split between training and test.

Task 1 - Build a Renewal Prediction Model: Build a model that uses the available data to predict whether or not a unit/resident renews. Discuss any preprocessing steps.

Task 2 - Model Evaluation: Identify a metric that demonstrates your model's performance compared to a random baseline. Output a csv with predictions for each resident in the test dataset. The csv should have three mandatory columns: resident_id, a unit_number and renewal_decision, please add probability_of_renewal if your model allows. We will run our internal eval on the test set.

Task 3 - Reason Categorization: Using the gte-base sentence transformer model, train a model to categorize the reasons tenants do not renew their leases. Analyze the reasons given by tenants and come up with appropriate categories that encapsulate the range of reasons for not renewing. Discuss how you defined these categories and any patterns or insights observed from the available data.

Presentation of Work: Prepare a short presentation highlighting the high-level approach and results for the research deep-dive. Have a Jupyter notebook ready - which will be part of the submission - and which will be used during the research deep-dive. Make sure the notebook contains clean and well documented code.