CrackML by @ml.with.umang
Interview questions / ML System Design
ML System Design interview question

Design Hashtag-to-Video Relevance

Design a system that predicts whether a hashtag is relevant to a video and uses that signal in product ranking.

hardsystem designEvidence 34/1001 source reportTikTok

The 60-second answer

Define relevance and downstream use first: search retrieval, recommendation, spam prevention, or hashtag quality can require different thresholds. Build labels from creator choices plus human review/engagement carefully; encode video frames, ASR/OCR/caption, hashtag text, and contextual metadata.

Build the answer in this order

1
Frame the problem

Define relevance and downstream use first: search retrieval, recommendation, spam prevention, or hashtag quality can require different thresholds.

2
Design the data path

Build labels from creator choices plus human review/engagement carefully; encode video frames, ASR/OCR/caption, hashtag text, and contextual metadata.

3
Choose the modeling stack

Use dual-encoder retrieval-style similarity or richer cross-modal matching, calibrated per hashtag frequency and language.

4
Serve, evaluate, iterate

Evaluate precision/recall by head/tail hashtag, language/content type, online search/recommendation impact, and abuse/adversarial tagging.

A useful interview mental model

This is the shape of a strong answer—not a script to memorize.

01Requirements
02Data
03Model / Retrieval
04Serving
05Monitor

Senior-level signal

  • Senior answers discuss popularity bias, weak-label contamination, and tail-hashtag cold start.
  • Separate semantic relevance from policy/safety eligibility and monitor both independently.

What the interviewer is really testing

Product framing, data design, modeling choices, serving constraints, reliability, evaluation, and explicit trade-offs.

Likely follow-up questions

What changes at 10× traffic or data volume?
Which failure mode would you monitor first in production?
How would you evaluate this offline and online before rollout?

Common weak-answer patterns

  • Jumping to a model before defining the product contract.
  • Listing components without bottlenecks, metrics, or failure handling.
  • Ignoring data quality, serving latency, monitoring, and iteration.