Design a customer feedback analytics system that turns millions of reviews, tickets and survey responses into themes and trends.

Design details

1. Scale and cost: millions of items, so use batch APIs or a fine-tuned small model for per-item tagging; reserve the LLM for cluster labelling and summaries.
2. Taxonomy management: automated clustering discovers new themes; humans approve and merge them into a stable taxonomy (so trends are comparable over time).
3. Emerging issue detection: spikes in new clusters or theme volume (e.g. "app crashes after update 5.2"), with alerts to product teams.
4. Trustworthy summaries: every insight links to representative quotes and counts computed by code (no LLM-made numbers).
5. Multilingual: analyse in the original language or translate; check sentiment accuracy per language.
6. NL Q&A: text-to-SQL over aggregates + RAG over quotes ("Why are users in Tamil Nadu unhappy with delivery?").

Evaluation: tagging accuracy on a labelled sample, cluster coherence (human rating), alert precision, product team usage.

Related

• Previous: Q40. Design a system where a fraud-detection ML model flags transactions and an LLM helps investigators understand and resolve cases.
• Next: Q42. Design a self-service fine-tuning platform for internal teams (upload data → train → evaluate → deploy).
• Machine Learning
• RAG Interview Questions