RAG on Support Tickets

In this example we'll index customer support tickets and query them to find the most similar ticket to a new one.

Let's create our first index where we'll add imaginary support tickets.

from opperai import Opper
from opperai.types import Document

# If not provided, the client will use the OPPER_API_KEY environment variable.
opper = Opper()
index = opper.indexes.create('support-tickets')

import Client from 'opperai';
import {Document} from 'opperai/types';

// If not provided, the client will use the OPPER_API_KEY environment variable.
const client = new Client();

const index = await client.indexes.create('support-tickets');

Now let's add some support tickets to our index.

tickets = [
     {
        "title": "Issue with my account",
        "description": "I cannot log in to my account",
        "status": "open",
        "id": 1,
    },
    {
        "title": "Payment issue",
        "description": "I cannot pay for my subscription",
        "status": "open",
        "id": 2,
    },
    {
        "title": "Feature request",
        "description": "I would like to request a new feature",
        "status": "closed",
        "id": 3,
    },
    {
        "title": "Slow response time",
        "description": "The response time for queries is too slow",
        "status": "open",
        "id": 4,
    },
    {
        "title": "Data sync error",
        "description": "There is an error syncing data across devices",
        "status": "open",
        "id": 5,
    },
    {
        "title": "App crash on launch",
        "description": "The app crashes every time I try to launch it",
        "status": "open",
        "id": 6,
    }, 
]

for ticket in tickets:
    index.add(Document(
        content=ticket["title"] + ' ' + ticket["description"],
        metadata={
            "status": ticket["status"],
            "id": ticket["id"],
        }
    ))

const tickets = [
    {
        title: 'Issue with my account',
        description: 'I cannot log in to my account',
        status: 'open',
        id: '1',
    },
    {
        title: 'Payment issue',
        description: 'I cannot pay for my subscription',
        status: 'open',
        id: '2',
    },
    {
        title: 'Feature request',
        description: 'I would like to request a new feature',
        status: 'closed',
        id: '3',
    },
    {
        title: 'Slow response time',
        description: 'The response time for queries is too slow',
        status: 'open',
        id: '4',
    },
    {
        title: 'Data sync error',
        description: 'There is an error syncing data across devices',
        status: 'open',
        id: '5',
    },
    {
        title: 'App crash on launch',
        description: 'The app crashes every time I try to launch it',
        status: 'open',
        id: '6',
    },
];

// Add each ticket to the index. Add status to metadata so we can filter by it later.
for (const ticket of tickets) {
    let doc:Document = {
        content: ticket.title + ' ' + ticket.description,
        metadata: {
            "status": ticket.status,
            "id": ticket.id,
        }
    }
    await client.indexes.add(index,doc);
}

Now, let's imagine that another customer has an issue with their account. We can query our index to find the most similar support ticket

query = 'Issue with my account'
results = index.query(query, 2)

print(results[0].content)
# 'Issue with my account I cannot log in to my account'
print(results[0].metadata)
# { status: 'open', id: '1' }

const query = 'Issue with my account';
const results = await client.indexes.retrieve(index, query, 2);

console.log(results[0].content);
// 'Issue with my account I cannot log in to my account'
console.log(results[0].metadata);
// { status: 'open', id: '1' }

You can also filter by metadata, for example, to only retrieve open tickets.

open_results = index.query(query, 1, [{'key': 'status', 'operation': '=', 'value': 'open'}])

print(open_results[0].content)
# 'Issue with my account I cannot log in to my account'

const open_results = await client.indexes.retrieve(index, query, 1, [{ key: 'status', operation: '=', value: 'open' }]);

console.log(open_results[0].content);
// 'Issue with my account I cannot log in to my account'

That's it! Indexes are a powerful tool at the core of many AI applications. When combined with functions, they can give knowledge about your data to LLMs and enable a lot of interesting use cases.