Posted 1h ago
Loading this job…
Data Science Intern - Model Evaluation and Experimentation
Posted 2h ago0 views
A short, sharp brief
Two months is not long, so this internship has exactly one goal: give MailerMen an evaluation harness the whole team trusts. Right now every model on the board is judged by whoever trained it, inside whatever notebook they happened to open that week.
The build
You define the offline metric set for the job matching model, wire up a held-out slice that respects time order instead of shuffling leakage into the results, and produce a single command that reruns every evaluation and prints a comparison table. Alongside that you run slice analysis: does the model do worse on internships than senior roles, on smaller cities than Bengaluru, on postings under a hundred words. Those answers change what we build next.
Working style
Fully remote inside India, with a short standup three mornings a week and one longer review on Friday. You report to our data lead and present findings to the founding team in your final week. A good fit if you are a final-year student or recent graduate who enjoys arguing about metrics and writes clearly, because half of this role is explaining what a number means to someone who did not train the model.
Responsibilities
- Define the offline metric set used to judge the job matching model and get the team to agree on it
- Build a time-ordered held-out split that removes the leakage in the current shuffled evaluation
- Write a single command that reruns every evaluation and prints a versioned comparison table
- Run slice analysis across role seniority, city tier and posting length to find where the model is weakest
- Document each metric in plain language so product and growth can read results unaided
- Present your findings to the founding team in the final week of the internship
Requirements
- Final-year student or recent graduate with a statistics, data science or analytics background
- Python and pandas, plus enough SQL to pull your own data without waiting on anyone
- Knows ranking metrics such as precision at k, recall at k and NDCG, or will learn them in week one
- Writes clearly, since explaining a metric is half the job
- Available for three short morning standups each week
Skills
Benefits
- Monthly stipend for a focused two-month engagement
- Direct reporting line to the data lead with a weekly review session
- A closing presentation to the founders, rarely on offer this early in a career
- Completion certificate and a recommendation letter