Upload 37 files

.env

@@ -1,7 +1,7 @@
 FASTAPI_CONFIG=production
 SECRET=zu=*nck@&r26rsa$2qv8a7g-p$slw79ym81mylj6s**w(da6&#
-OPENAI_API_KEY=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
-
+OPENAI_API_KEY=sk-proj-GYY7hx5SM02oDz5lhrK0T3BlbkFJegeuu9bXdNONd3Ds1rqY
+GOOGLE_PLACES_API_KEY=AIzaSyDvND-yu2b3p2FcO5aS7YhuyFI4dsXRBts
 #DATABASE_USER=hectool_ai_filter
 #DATABASE_PASSWORD=nixtz3orhepndjw4m1D4
 #DATABASE_HOST=localhost

project/bot/documents.py

@@ -3,33 +3,33 @@ import pandas as pd
 
 with open('../../data.json', 'r') as f:
     data = json.load(f)
-
+#
 chunks = []
-for post in data:
-    post_text = post['text']
-    comments: list[dict] = post['comments']
-    comments_str = ''
-    for i, comment in enumerate(comments):
-        comment_text = list(comment.keys())[0]
-        replies = comment[comment_text]
-        reply_str = 'Replies:\n'
-        for j, reply in enumerate(replies):
-            if j + 1 == len(replies):
-                reply_str += f'     • {reply}'
-            else:
-                reply_str += f'     • {reply}\n'
-        comments_str += f'{i + 1}. {comment_text}\n'
-        if replies:
-            comments_str += f'{reply_str}\n'
-
-    chunk = f"Post: {post_text}\n"
-    if comments:
-        chunk += f'Comments:\n{comments_str}'
-    chunks.append(chunk)
+# for post in data:
+#     post_text = post['text']
+#     comments: list[dict] = post['comments']
+#     comments_str = ''
+#     for i, comment in enumerate(comments):
+#         comment_text = list(comment.keys())[0]
+#         replies = comment[comment_text]
+#         reply_str = 'Replies:\n'
+#         for j, reply in enumerate(replies):
+#             if j + 1 == len(replies):
+#                 reply_str += f'     • {reply}'
+#             else:
+#                 reply_str += f'     • {reply}\n'
+#         comments_str += f'{i + 1}. {comment_text}\n'
+#         if replies:
+#             comments_str += f'{reply_str}\n'
+#
+#     chunk = f"Post: {post_text}\n"
+#     if comments:
+#         chunk += f'Comments:\n{comments_str}'
+#     chunks.append(chunk)
+# #
+# df = pd.DataFrame({"chunks": chunks})
+# df.to_csv('chunks_javea.csv', index=False)
 #
-df = pd.DataFrame({"chunks": chunks})
-df.to_csv('chunks_javea.csv', index=False)
-
 # for post in data:
 #     post_text = post['text']
 #     comments: list[dict] = post['comments']
@@ -54,3 +54,30 @@ df.to_csv('chunks_javea.csv', index=False)
 
 # df = pd.DataFrame({"chunks": chunks})
 # df.to_csv('chunks_javea_raw.csv', index=False)
+
+# import requests
+# import json
+#
+# # URL для запроса к API
+# url = "https://places.googleapis.com/v1/places:searchText"
+#
+# # API ключ (замените 'API_KEY' на ваш реальный API-ключ)
+# api_key = "AIzaSyAcfgJAVTK1gwC9tmeKg-wcVWCy5CdieW4"
+#
+# # Заголовки запроса
+# headers = {
+#     "Content-Type": "application/json",
+#     "X-Goog-Api-Key": api_key,
+#     "X-Goog-FieldMask": "places.displayName,places.formattedAddress,places.priceLevel"
+# }
+#
+# # Тело запроса
+# data = {
+#     "textQuery": "Spicy Vegetarian Food in Sydney, Australia"
+# }
+#
+# # Отправка POST запроса
+# response = requests.post(url, headers=headers, data=json.dumps(data))
+#
+# # Вывод ответа от сервера
+# print(response.text)

project/bot/openai_backend.py

@@ -1,10 +1,14 @@
 import asyncio
+import json
+import re
 from typing import List, Dict
 import faiss
+import httpx
 import numpy as np
 import pandas as pd
 from sqlalchemy.ext.asyncio import AsyncSession
 from starlette.websockets import WebSocket
+from transformers import pipeline
 
 from project.bot.models import MessagePair
 from project.config import settings
@@ -12,6 +16,7 @@ from project.config import settings
 
 class SearchBot:
     chat_history = []
+
     # is_unknown = False
     # unknown_counter = 0
 
@@ -20,33 +25,70 @@ class SearchBot:
             memory = []
         self.chat_history = memory
 
-    async def _summarize_user_intent(self, user_query: str) -> str:
-        chat_history_str = ''
-        chat_history = self.chat_history[-self.unknown_counter * 2:]
-        for i in chat_history:
-            if i['role'] == 'user':
-                chat_history_str += f"{i['role']}: {i['content']}\n"
-        messages = [
-            {
-                'role': 'system',
-                'content': f"{settings.SUMMARIZE_PROMPT}\n"
-                           f"Chat history: ```{chat_history_str}```\n"
-                           f"User query: ```{user_query}```"
-            }
-        ]
-        response = await settings.OPENAI_CLIENT.chat.completions.create(
-            messages=messages,
-            temperature=0.1,
-            n=1,
-            model="gpt-3.5-turbo-0125"
-        )
-        user_intent = response.choices[0].message.content
-        return user_intent
-
     @staticmethod
     def _cls_pooling(model_output):
         return model_output.last_hidden_state[:, 0]
 
+    @staticmethod
+    async def enrich_information_from_google(search_word: str) -> str:
+        url = "https://places.googleapis.com/v1/places:searchText"
+        headers = {
+            "Content-Type": "application/json",
+            "X-Goog-Api-Key": settings.GOOGLE_PLACES_API_KEY,
+            "X-Goog-FieldMask": "places.shortFormattedAddress,places.websiteUri,places.internationalPhoneNumber,"
+                                "places.googleMapsUri,places.photos"
+        }
+        data = {
+            "textQuery": f"{search_word} in Javea",
+            "languageCode": "nl",
+            "maxResultCount": 1,
+
+        }
+        async with httpx.AsyncClient() as client:
+            response = await client.post(url, headers=headers, content=json.dumps(data))
+            place_response = response.json()
+            place_response = place_response['places'][0]
+        photo_name = place_response.get('photos')
+        photo_uri = None
+        if photo_name:
+            async with httpx.AsyncClient() as client:
+                response = await client.get(
+                    f'https://places.googleapis.com/v1/{photo_name[0]["name"]}/media?maxWidthPx=350&key={settings.GOOGLE_PLACES_API_KEY}')
+                photo_response = response.json()
+            photo_uri = photo_response.get('photoUri')
+        google_maps_uri = place_response.get('googleMapsUri')
+        phone_number = place_response.get('internationalPhoneNumber')
+        formatted_address = place_response.get('shortFormattedAddress')
+        website_uri = place_response.get('websiteUri')
+        if not google_maps_uri:
+            return search_word
+        enriched_word = f'<a class="extraDataLink" href="{google_maps_uri}" target="_blank">{search_word}</a><div class="tooltip-elem">'
+        if photo_uri:
+            enriched_word += f'<img src="{photo_uri}" alt="Image" class="tooltip-img">'
+        if formatted_address:
+            enriched_word += f'<p><a href="{google_maps_uri}" target="_blank">{formatted_address}</a></p>'
+        if website_uri:
+            enriched_word += f'<p><a href="{website_uri}">Google Maps URI</a></p>'
+        if phone_number:
+            phone_str = re.sub(r' ', '', phone_number)
+            enriched_word += f'<p><a href="tel:{phone_str}">Phone number</a></p>'
+        enriched_word += f"</div>"
+        return enriched_word
+
+    async def analyze_full_response(self) -> str:
+        assistant_message = self.chat_history.pop()['content']
+        nlp = pipeline("ner", model=settings.NLP_MODEL, tokenizer=settings.NLP_TOKENIZER, grouped_entities=True)
+        ner_result = nlp(assistant_message)
+        analyzed_assistant_message = assistant_message
+        for entity in ner_result:
+            if entity['entity_group'] in ("LOC", "ORG", "MISC") and entity['word'] != "Javea":
+                start = entity['start']
+                end = entity['end']
+                enriched_information = await self.enrich_information_from_google(entity['word'])
+                analyzed_assistant_message = analyzed_assistant_message[
+                                             :start] + f"{enriched_information}" + analyzed_assistant_message[end:]
+        return "ENRICHED:" + analyzed_assistant_message
+
     async def _convert_to_embeddings(self, text_list):
         encoded_input = settings.INFO_TOKENIZER(
             text_list, padding=True, truncation=True, return_tensors="pt"
@@ -57,7 +99,7 @@ class SearchBot:
 
     @staticmethod
     async def _get_context_data(user_query: list[float]) -> list[dict]:
-        radius = 30
+        radius = 4
         _, distances, indices = settings.FAISS_INDEX.range_search(user_query, radius)
         indices_distances_df = pd.DataFrame({'index': indices, 'distance': distances})
         filtered_data_df = settings.products_dataset.iloc[indices].copy()
@@ -83,7 +125,6 @@ class SearchBot:
         else:
             content = settings.EMPTY_PROMPT
         user_message = {"role": 'user', "content": query}
-
         self.chat_history.append(user_message)
         messages = [
             {
@@ -122,7 +163,8 @@ class SearchBot:
         try:
             async for chunk in self._rag(context, query, session, country):
                 await websocket.send_text(chunk)
-            # await websocket.send_text('finish')
+            analyzing = await self.analyze_full_response()
+            await websocket.send_text(analyzing)
         except Exception:
             await self.emergency_db_saving(session)
 

project/bot/templates/home.html

@@ -12,8 +12,8 @@
     <link href="https://cdn.jsdelivr.net/npm/[email protected]/dist/css/bootstrap.min.css" rel="stylesheet"
           integrity="sha384-rbsA2VBKQhggwzxH7pPCaAqO46MgnOM80zW1RWuH61DGLwZJEdK2Kadq2F9CUG65" crossorigin="anonymous">
     <!--     Connect style.css -->
-<!--     <link rel="stylesheet" href="{{ url_for('static', path='/css/style.css') }}"> -->
-   <link rel="stylesheet" href="https://brestok-javeaai.hf.space/static/css/style.css">
+    <link rel="stylesheet" href="{{ url_for('static', path='/css/style.css') }}">
+    <!--    <link rel="stylesheet" href="../../../static/css/style.css">-->
 </head>
 
 <div class="container-fluid px-0">
@@ -44,9 +44,7 @@
                 <div class="chat-body" id="chatBody">
                     <div class="message">
                         <div class="bot_message mt-4 py-3 px-4 rounded-4 w-75 ms-3">
-                            Hallo! Ik ben een virtuele assistent, een helper voor Nederlanders die naar Javea zijn
-                            gemigreerd. Ik weet alles over deze regio en zal je helpen om je aan te passen in je nieuwe
-                            thuis. Waarmee kan ik je helpen?
+                            Hallo! Ich bin der virtuelle Assistent Alcolm AI. Sie können sich jederzeit mit Fragen zum Alcolm-Software an mich wenden. Ich helfe Ihnen gerne weiter!
                         </div>
                     </div>
                 </div>
@@ -64,6 +62,7 @@
     </div>
 </div>
 
+
 </html>
 <script src="https://cdn.jsdelivr.net/npm/[email protected]/dist/js/bootstrap.bundle.min.js"
         integrity="sha384-kenU1KFdBIe4zVF0s0G1M5b4hcpxyD9F7jL+jjXkk+Q2h455rYXK/7HAuoJl+0I4"
@@ -74,9 +73,9 @@
 <script src="https://code.jquery.com/ui/1.12.1/jquery-ui.js"></script>
 <script src="https://kit.fontawesome.com/d4ffd37f75.js" crossorigin="anonymous"></script>
 <script src="https://cdn.jsdelivr.net/npm/marked/marked.min.js"></script>
-<script type="text/javascript" src="https://brestok-javeaai.hf.space/static/js/main.js"></script>
-<script type="text/javascript" src="https://brestok-javeaai.hf.space/static/js/utils.js"></script>
-<script type="text/javascript" src="https://brestok-javeaai.hf.space/static/js/ws.js"></script>
-<!-- <script type="text/javascript" src="{{ url_for('static', path='/js/main.js') }}"></script> -->
-<!-- <script type="text/javascript" src="{{ url_for('static', path='/js/utils.js') }}"></script> -->
-<!-- <script type="text/javascript" src="{{ url_for('static', path='/js/ws.js') }}"></script> -->
+<!--<script type="text/javascript" src="../../../static/js/main.js"></script>-->
+<!--<script type="text/javascript" src="../../../static/js/utils.js"></script>-->
+<!--<script type="text/javascript" src="../../../static/js/ws.js"></script>-->
+<script type="text/javascript" src="{{ url_for('static', path='/js/main.js') }}"></script>
+<script type="text/javascript" src="{{ url_for('static', path='/js/utils.js') }}"></script>
+<script type="text/javascript" src="{{ url_for('static', path='/js/ws.js') }}"></script>

project/config.py

@@ -6,7 +6,7 @@ from openai import AsyncOpenAI
 import pathlib
 from functools import lru_cache
 from environs import Env
-from transformers import AutoModel, AutoTokenizer
+from transformers import AutoModel, AutoTokenizer, AutoModelForTokenClassification
 import torch
 
 env = Env()
@@ -21,6 +21,9 @@ class BaseConfig:
     INFO_TOKENIZER = AutoTokenizer.from_pretrained(MODEL_NAME)
     OPENAI_CLIENT = AsyncOpenAI(api_key=os.getenv('OPENAI_API_KEY'))
     FAISS_INDEX = faiss.read_index(str(BASE_DIR / 'faiss_javea.index'))
+    NLP_MODEL_NAME = 'Babelscape/wikineural-multilingual-ner'
+    NLP_TOKENIZER = AutoTokenizer.from_pretrained("Babelscape/wikineural-multilingual-ner")
+    NLP_MODEL = AutoModelForTokenClassification.from_pretrained("Babelscape/wikineural-multilingual-ner")
 
 
 class DevelopmentConfig(BaseConfig):
@@ -33,24 +36,21 @@ class ProductionConfig(BaseConfig):
                    f"{env('DATABASE_HOST')}:" \
                    f"{env('DATABASE_PORT')}/" \
                    f"{env('DATABASE_NAME')}"
-    PROMPT = "Je bent een expert in de regio Javea in Italië, die alles weet om mensen te helpen die migreren van " \
-             "Nederland naar Spanje. Je taak is om mensen te helpen zich te vestigen in de nieuwe stad. Gebruik " \
-             "kennis uit je vorige antwoord (voornamelijk uit opmerkingen) om een informatief antwoord " \
-             "te geven op de gebruikersvraag. Vermeld nooit dat je kennis haalt uit posts of opmerkingen. Spreek " \
-             "vanuit jezelf."
+    PROMPT = "Je bent een expert in de Spaanse regio Javea die alles weet over hoe je mensen kunt helpen die vanuit " \
+             "Nederland naar Spanje migreren. Jouw taak is om mensen te helpen zich te vestigen in een nieuwe stad. " \
+             "Gebruik de kennis die je in je vorige antwoord hebt opgedaan (meestal uit opmerkingen) om een " \
+             "informatief antwoord te geven op de vraag van de gebruiker. Concentreer je in je antwoorden op de namen " \
+             "van de locaties. Vermeld nooit dat je kennis afkomstig is van berichten of opmerkingen. Spreek namens " \
+             "jezelf. "
     EMPTY_PROMPT = "Je bent een expert in Javea aan de Costa Blanca in Spanje, met uitgebreide kennis om Nederlanders " \
                    "te helpen die naar deze regio verhuizen. Je taak is om mensen te helpen zich thuis te voelen in " \
                    "hun nieuwe stad. Gebruik je kennis over deze regio maar informatieve antwoorden te " \
                    "geven op de vragen van gebruikers."
-    SUMMARIZE_PROMPT = "Study the user's requests, paying special attention to the specific mentioned wishes when " \
-                       "choosing a house. Combine these details into a single query that reflects all the user's " \
-                       "needs. Formulate your answer as if you were a user, clearly and concisely stating the " \
-                       "requirements. Make sure that all relevant user wishes are indicated in your response. "
+    GOOGLE_PLACES_API_KEY = env('GOOGLE_PLACES_API_KEY')
 
     def __init__(self):
         if torch.cuda.is_available():
             device = torch.device("cuda")
-
         else:
             device = torch.device("cpu")
         self.device = device

static/css/style.css

@@ -11,6 +11,8 @@ body {
 .chat-header {
     background-color: #f1f1f1;
     border-bottom: 1px solid #bdbdbd;
+    /*z-index: 10;*/
+    /*position: relative;*/
 }
 
 .chat-body {
@@ -79,3 +81,39 @@ body {
     padding: 20px;
     box-shadow: 0 4px 8px rgba(0, 0, 0, 0.1);
 }
+
+.extraDataLink {
+    color: #105ead;
+    text-decoration: none;
+    position: relative;
+}
+
+.extraDataLink:hover {
+    color: #1164b7;
+    text-decoration: none;
+    z-index: 150;
+}
+
+.tooltip-elem {
+    position: absolute;
+    background-color: black;
+    color: white;
+    font-size: 12px;
+    padding: 8px;
+    border-radius: 4px;
+    z-index: 1000;
+    display: none;
+}
+.extraDataLink + .tooltip-elem {
+    position: absolute;
+}
+
+.tooltip-img{
+    border-radius: 4px;
+    margin-bottom: 5px;
+}
+
+.tooltip-elem a{
+    text-decoration: none;
+    color: #008dff;
+}

static/js/main.js

@@ -21,7 +21,6 @@ function adjustChatBodyHeight() {
     const chatHeader = document.getElementById('chatHeader')
     const chatFooterHeight = getTotalHeight(chatFooter);
     const chatHeaderHeight = getTotalHeight(chatHeader);
-    const openButtonWindow = getTotalHeight(openButton)
     const viewportHeight = window.innerHeight - chatHeaderHeight - chatFooterHeight;
     chatBody.style.height = viewportHeight + 'px';
 }
@@ -29,7 +28,7 @@ function adjustChatBodyHeight() {
 function openChatBotWindow() {
     let lastScrollHeight = chatBody.scrollHeight;
     const uuid = generateUUID()
-    socket = new WebSocket(`wss://brestok-javeaai.hf.space/ws/${uuid}`);
+    socket = new WebSocket(`ws://127.0.0.1:8000/ws/${uuid}`);
     socket.onclose = (event) => console.log('WebSocket disconnected', event);
     socket.onerror = (error) => {
         alert('Something was wrong. Try again later.')
@@ -46,7 +45,13 @@ function openChatBotWindow() {
             const botMessages = document.querySelectorAll('.bot_message');
             botMessage = botMessages[botMessages.length - 1];
         } else {
-            botMessage.innerHTML = marked.parse(event.data)
+            const aiMessage = event.data
+            if (aiMessage.startsWith('ENRICHED:')) {
+                botMessage.innerHTML = aiMessage.substring(9)
+                enrichAIResponse(botMessage)
+            } else {
+                botMessage.innerHTML = marked.parse(aiMessage)
+            }
         }
     }
 }
@@ -61,4 +66,4 @@ openButton.addEventListener('click', function () {
     chatbotWindow.style.visibility = "visible";
     openButton.style.display = 'none'
     openChatBotWindow();
-});
+});

static/js/utils.js

@@ -5,6 +5,7 @@ function createNewMessage(text, type) {
     const message = document.createElement('div')
     message.className = 'message'
     let content = document.createElement('div')
+    content.style.whiteSpace = "pre-line"
     if (type === 'bot') {
         content.className = 'bot_message mt-4 py-3 px-4 rounded-4 w-75 ms-3'
     } else {
@@ -45,4 +46,42 @@ textInput.addEventListener("keydown", function (event) {
     }
 });
 
-sendButton.addEventListener("click", sendMessageToServer);
+sendButton.addEventListener("click", sendMessageToServer);
+
+
+function enrichAIResponse(botMessageElement) {
+    const links = botMessageElement.querySelectorAll('.extraDataLink');
+
+    links.forEach(link => {
+        const tooltip = link.nextElementSibling;
+        link.addEventListener('mouseenter', function () {
+            tooltip.style.display = 'block';
+            const tooltipWidth = parseInt(link.offsetWidth) * 2
+            tooltip.style.width = tooltipWidth + 'px'
+
+            const tooltipImg = tooltip.querySelector('.tooltip-img')
+            if (tooltipImg) {
+                tooltipImg.width = tooltipWidth - 16
+            }
+            const linkRect = this.getBoundingClientRect();
+            if (linkRect.top < tooltip.offsetHeight + 4) {
+                tooltip.style.top = (window.scrollY + linkRect.bottom + 4) + 'px';
+            } else {
+                tooltip.style.top = (window.scrollY + linkRect.top - tooltip.offsetHeight - 4) + 'px';
+            }
+            tooltip.style.left = (linkRect.left + (linkRect.width / 2) - (tooltipWidth / 2)) + 'px';
+        });
+
+        link.addEventListener('mouseleave', function () {
+            setTimeout(() => {
+                if (!tooltip.matches(':hover')) {
+                    tooltip.style.display = 'none';
+                }
+            }, 300)
+        });
+
+        tooltip.addEventListener('mouseleave', function () {
+            this.style.display = 'none';
+        });
+    });
+}