Skip to content
Selected work

AI Ticketing System

Duplicate detection over a live support queue

One recurring issue arrives as a dozen differently-worded tickets, and keyword search catches none of them.

Context
Internship — Sole Developer
Stack
Python · ConnectWise Manage · Ollama · nomic-embed · Cosine similarity

The problem

A managed IT services team was handling the same underlying issue several times over, because a recurring problem arrives as a dozen separately-worded tickets. Keyword search doesn't catch it, nothing matches literally.

What I built

Every ticket is embedded with nomic-embed on a local LLM stack via Ollama, so nothing leaves the company's infrastructure, and cosine similarity over those vectors surfaces near-duplicates and recurring patterns that share no keywords. On top of the vector layer sits a narrative generator: instead of a wall of tickets, the team gets a written summary of what is actually happening across the queue.

How it fits together

  1. ConnectWise queuelive support tickets
  2. On their own hardware — no ticket text leaves the network
    1. nomic-embedone vector per ticket, via Ollama
    2. Cosine similaritymatches meaning, not keywords
    3. Duplicate clustersone issue, a dozen wordings
  3. Narrative summarywhat is actually happening across the queue
Pipeline: tickets from the ConnectWise queue are embedded with nomic-embed through a local Ollama stack, cosine similarity over those vectors clusters near-duplicates that share no keywords, and a generator writes the queue back as a narrative summary. Embedding, similarity and clustering all run on the company's own hardware, so no ticket text leaves the network.