Dailyhunt · GenAI · Support
Customer support RAG tool
Unified, grounded LLM replies across four feedback channels — vernacular and Hinglish —
so humans only handle the hard cases.
Problem
- CX handled feedback across four disconnected channels — Play Store, App Store, email, in-app reviews
- Manual read → bucket → draft reply → escalate — no unified view of inbound volume
- Slow, inconsistent tone, and hard to scale across 5+ products
What I built
- RAG system: retrieve the right template / FAQ by issue bucket, then generate a grounded reply
- LLM generation in vernacular languages and Hinglish — not a generic canned template
- Defined eval criteria from scratch (retrieval accuracy + tone quality)
- Iterated on real feedback loops — not a one-shot ship
Outcome
- ~90% of support feedback resolved automatically
- 120+ hours / month saved for the CX team
- ~₹20L impact across 5+ products
- Human team focused on complex cases that need judgment