Expect ranking design under competing objectives (ranking supplier listings for a price-sensitive shopper while balancing relevance, price, and quality), cost-and-guardrail framing (reducing returns without hurting orders, framed end to end as ML and experimentation), and evaluation rigour (off-policy evaluation and selection bias when you only have logged data from the current production model). Interviewers reward candidates who reason about multiple objectives, real business costs, and the biases baked into logged data rather than optimising a single metric. Prepare ranking, experiment design, and off-policy evaluation grounded in marketplace scenarios.
Recruitment process
- Recruiter screen and technical pre-screen
- ML fundamentals and coding round
- ML design and experimentation round
- Hiring-manager and behavioural round, then offer
Interview rounds
- Round 1 (45-60 min): ML fundamentals and coding round
- Round 2 (60 min): ranking and ML design round with multiple objectives
- Round 3 (45-60 min): experimentation and off-policy-evaluation round
- Round 4 (45 min): behavioural and hiring-manager round
Multi objective and unbiased framework
- Rank under competing objectives, frame problems around real costs and guardrails, and evaluate rigorously against selection bias in logged data
Practice questions
- 01Practise
Design a pipeline that computes supplier catalogue-quality scores from listing, order, and return events. How do you handle high return rates skewing the signal?
IntenseMeesho data-engineer rounds probe multi-event scoring pipelines and handling of noisy signals like returns in a social-commerce context. - 02Practise
Write SQL to find, per category, the suppliers whose order volume grew fastest month-over-month, excluding suppliers with fewer than 100 orders.
StandardMeesho tests window-function growth metrics with LAG and threshold filtering for supplier analytics. - 03Practise
How would you design a cost-efficient data platform for a value-focused marketplace where controlling warehouse and compute spend is a hard constraint?
StandardMeesho is cost-conscious; rounds probe partition pruning, storage-format choices, and compute-cost trade-offs in platform design. - 04Practise
Design a model to rank supplier listings for a value-conscious shopper where price sensitivity is high. How do you balance relevance, price, and quality?
IntenseMeesho data-scientist rounds test ranking under multiple objectives: relevance, price sensitivity, and quality signals for a value marketplace. - 05Practise
You want to reduce returns without hurting orders. Frame this as an ML and experimentation problem end to end.
StandardMeesho probes problem framing tied to a real cost (returns) with a guardrail (orders), plus sound experiment design. - 06Practise
Explain how you'd evaluate a recommendation model offline when you only have logged data from the current production model (selection bias).
IntenseMeesho data-scientist rounds probe off-policy evaluation and selection bias in logged recommendation data.
