Google 2026 New Grad San Jose Onsite: Graph Power and Next-Word Prediction

Role: Software Engineer, New Grad
Result: Awaiting recruiter call
Onsite date: January 29, 2026
Location: San Jose

Coding 1: graph power levels

Nodes form a graph. Torch nodes begin with power 16, and power decays with distance along wires. Compute node power levels.

The candidate used multi-source BFS from the torches to compute distances. They proactively discussed cycles; the interviewer expected cycle detection, although the completed BFS would already avoid infinite traversal with a visited structure.

The source never defines the decay formula, the required output, edge direction, unreachable-node behavior, or how multiple torches combine. Those are core rules, so no exact practice problem is linked.

Coding 2: next-word predictor

Process a word sequence and track how often each word follows another word. Given a word, return its most frequent successor.

The candidate initially proposed a priority queue. The interviewer clarified that the dataset would be processed once as a batch, so the candidate used a map of maps: word -> {nextWord -> count}.

Practice: Most-Frequent Next-Word Predictor

The candidate described this as a very small NLP preprocessing task. At the time of the post, they had scheduled a recruiter call but did not yet know the final outcome.