Google MLE IC4 Offer: Correctable Stock Prices, Streamed Triplets, and ML Domain

Role: Machine Learning Engineer, IC4 / PhD New Grad
Result: Offer
Report time: June 2025

The candidate had applied to Google for several years before receiving this interview and ultimately accepted an offer.

Coding 1: correctable stock-price tracker

Track one stock from a live feed of (timestamp, price) updates and expose its highest, lowest, and most recent price.

The feed may correct historical data:

• (timestamp, newPrice) replaces that timestamp's price.
• (timestamp) deletes that timestamp's price.

Implement a class that ingests already-parsed operations and maintains current, minimum, and maximum values.

Practice: Correctable Stock Price Tracker

Coding 2: streamed scalar triplets

Given distance d, ingest floating-point values one at a time. When three stored values are all within d of one another, return those three and remove them from memory.

Practice: Streamed Scalar Item Triplets

ML-domain interview

Topics included:

• ML training and optimizers
• What to try when training loss does not decrease
• Low-level ML frameworks
• Multi-GPU training and threading
• Vision models, including CNN and ViT
• Model changes made in past projects
• Language models
• Beam search
• Greedy, temperature-based, and other decoding strategies

Candidate advice and Team Match

The candidate recommended clarifying the most frequent operation, writing assumptions and pseudocode, explaining complexity, considering brute force briefly, then coding once the design is stable. During Team Match, candidates filled a preference form and could speak with one team at a time until both sides agreed.