Google·Software Engineer·Technical Phone Screen·Intermediate

IntermediatePrefer not to say

Jun 2026

Summary

Google coding interview with a pretty niche OCR-meets-digit-validation problem. The twist with 180-degree rotation made it way less straightforward than it first looked.

Questions Asked (1)

Build a validator that, given an image of handwritten digits, returns true if the number falls within [1, 200], and false otherwise. The catch: the card could be upright or rotated 180 degrees, and the validator must handle both orientations correctly.

Algorithms & Data StructuresTechnical Trade-offs

Author's notes

The base range check felt obvious and I coded it fast.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying requirements and assumptions, then propose a pipeline: digit detection/segmentation, orientation classification, digit recognition, and range validation. Discuss trade-offs between end-to-end deep learning and modular approaches, and how to handle ambiguity and errors.

Pro tip: Emphasize the importance of a confidence threshold for orientation and recognition; if confidence is low, you can try both orientations and use the recognition model's confidence to pick the correct one, which is more robust than a separate orientation classifier.

1. Clarify Requirements and Assumptions

Ask about image quality, number of digits, whether the number is always within a certain range, and if the rotation is exactly 180 degrees. Clarify if the validator should be robust to noise or if it's a controlled environment.

2. Design the Pipeline

Outline steps: preprocess image (grayscale, thresholding), detect and segment digits, determine orientation, recognize digits, and validate range. Discuss whether to use a single end-to-end model or separate modules.

3. Handle Orientation

Propose methods: train a classifier to predict orientation, or run recognition on both orientations and choose the one with higher confidence. Discuss data augmentation with rotated images during training.

4. Digit Recognition and Range Validation

Use a CNN or existing OCR for digit recognition. Convert recognized digits to an integer and check if it's in [1, 200]. Handle cases like leading zeros or multiple digits.

5. Evaluate and Iterate

Discuss metrics (accuracy, precision/recall), error analysis, and how to handle edge cases (e.g., ambiguous digits like 6/9). Consider ensemble methods or confidence thresholds to improve robustness.

Key Points to Mention

Data augmentation with 180-degree rotations to make the model orientation-invariant.
Trade-offs between end-to-end deep learning (simpler pipeline but needs more data) and modular approach (more interpretable, easier to debug).
Handling ambiguous digits (6 vs 9) by considering orientation context or using a confidence threshold.
Using a confidence-based approach for orientation: run recognition on both orientations and pick the one with higher confidence.
Preprocessing steps: normalization, deskewing, and noise removal to improve recognition accuracy.
Validation logic: ensure the recognized number is within [1, 200], and consider if the number could have leading zeros (e.g., '007' is 7).

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.