Skip to content
Mirun Kaushik
← All work

06Take-home assessment2026

Support Ticket Triage

Classifies and prioritises support tickets, with a semantic cache

My part
Solo
Status
Built
Stack
Node.js, Redis, BullMQ, pgvector, Hugging Face
Code
GitHub ↗

Fig. — How data moves through it

  1. 01Ticket
  2. 02BullMQRedis
  3. 03pgvector cachesimilar ticket seen?
  4. 04Hugging Face modelon cache miss(model call)
  5. 05Category + priority
Model call Rules, data, queues

01Problem

Support teams need incoming tickets sorted and prioritised without blocking on slow model calls.

02Approach

Process tickets asynchronously through job queues, and check a semantic cache first so the model isn’t called again for a question it has already answered.

03What I built

  1. 01A service that classifies and prioritises tickets with a Hugging Face model, processing them asynchronously through Redis-backed BullMQ job queues.
  2. 02A pgvector semantic cache that reuses responses for similar tickets to avoid repeated LLM calls.