query.py

import requests
import json
import re
from urllib.parse import quote

def extract_between_tags(text, start_tag, end_tag):
    start_index = text.find(start_tag)
    end_index = text.find(end_tag, start_index)
    return text[start_index+len(start_tag):end_index-len(end_tag)]

class VectaraQuery():
    def __init__(self, api_key: str, customer_id: int, corpus_ids: list):
        self.customer_id = customer_id
        self.corpus_ids = corpus_ids
        self.api_key = api_key
        self.conv_id = None

    def submit_query(self, query_str: str):
        corpora_key_list = [{
                'customer_id': str(self.customer_id), 'corpus_id': str(corpus_id), 'lexical_interpolation_config': {'lambda': 0.025}
            } for corpus_id in self.corpus_ids
        ]

        endpoint = f"https://api.vectara.io/v1/query"
        start_tag = "%START_SNIPPET%"
        end_tag = "%END_SNIPPET%"
        headers = {
            "Content-Type": "application/json",
            "Accept": "application/json",
            "customer-id": str(self.customer_id),
            "x-api-key": self.api_key,
            "grpc-timeout": "60S"
        }
        body = {
            'query': [
                { 
                    'query': query_str,
                    'start': 0,
                    'numResults': 50,
                    'corpusKey': corpora_key_list,
                    'context_config': {
                        'sentences_before': 2,
                        'sentences_after': 2,
                        'start_tag': start_tag,
                        'end_tag': end_tag,
                    },
                    'rerankingConfig':
                    {
                        'rerankerId': 272725718,
                        'mmrConfig': {
                            'diversityBias': 0.3
                        }
                    },
                    'summary': [
                        {
                            'responseLang': 'eng',
                            'maxSummarizedResults': 5,
                            'summarizerPromptName': 'vectara-experimental-summary-ext-2023-12-11-sml',
                            'chat': {
                                'store': True,
                                'conversationId': self.conv_id
                            },
                            'debug': True,
                        }
                    ]
                } 
            ]
        }
        
        response = requests.post(endpoint, data=json.dumps(body), verify=True, headers=headers)    
        if response.status_code != 200:
            print(f"Query failed with code {response.status_code}, reason {response.reason}, text {response.text}")
            return "Sorry, I'm experiencing an error. Please report this and try again later."

        res = response.json()

        top_k = 10
        summary = res['responseSet'][0]['summary'][0]['text']
        responses = res['responseSet'][0]['response'][:top_k]
        docs = res['responseSet'][0]['document']
        chat = res['responseSet'][0]['summary'][0]['chat']

        if chat['status'] != None:
            st_code = chat['status']
            print(f"Chat query failed with code {st_code}")
            if st_code == 'RESOURCE_EXHAUSTED':
                self.conv_id = None
                return 'Sorry, chat turns exceeds plan limit.'
            return 'Sorry, I'm experiencing an error. Please report this and try again later.'
        
        self.conv_id = res['responseSet'][0]['summary'][0]['chat']['conversationId']
        
        pattern = r'\[\d{1,2}\]'
        matches = [match.span() for match in re.finditer(pattern, summary)]

        # figure out unique list of references
        refs = []
        for match in matches:
            start, end = match
            response_num = int(summary[start+1:end-1])
            doc_num = responses[response_num-1]['documentIndex']
            metadata = {item['name']: item['value'] for item in docs[doc_num]['metadata']}
            text = extract_between_tags(responses[response_num-1]['text'], start_tag, end_tag)
            url = f"{metadata['url']}#:~:text={quote(text)}"
            if url not in refs:
                refs.append(url)

        # replace references with markdown links
        refs_dict = {url:(inx+1) for inx,url in enumerate(refs)}
        for match in reversed(matches):
            start, end = match
            response_num = int(summary[start+1:end-1])
            doc_num = responses[response_num-1]['documentIndex']
            metadata = {item['name']: item['value'] for item in docs[doc_num]['metadata']}
            text = extract_between_tags(responses[response_num-1]['text'], start_tag, end_tag)
            url = f"{metadata['url']}#:~:text={quote(text)}"
            citation_inx = refs_dict[url]
            summary = summary[:start] + f'[\[{citation_inx}\]]({url})' + summary[end:]

        return summary