Create app.py

app.py

@@ -0,0 +1,194 @@
+# app.py
+
+from flask import Flask, render_template, request, jsonify
+import faiss
+import numpy as np
+import json
+from sentence_transformers import SentenceTransformer
+from langchain.prompts import PromptTemplate
+from langchain_groq import ChatGroq
+import re
+from dotenv import load_dotenv
+load_dotenv()
+
+app = Flask(__name__)
+
+# Load Model and FAISS Index
+model = SentenceTransformer('./sentence-transformers_all-MiniLM-L6-v2')
+index = faiss.read_index("faiss_index.bin")
+groq_api_key = os.getenv('GROQ_API_KEY')
+model_name = "llama-3.3-70b-versatile"
+
+llm = ChatGroq(
+            temperature=0, 
+            groq_api_key=groq_api_key, 
+            model_name=model_name
+        )
+
+with open("metadata.json") as f:
+    metadata = json.load(f)
+
+
+def categorize_query(query):
+    """
+    Categorizes user queries into different types (greetings, small talk, unrelated, etc.).
+    """
+    query = query.lower().strip()
+
+    # Greetings
+    greeting_patterns = [
+        r"\bhello\b", r"\bhi\b", r"\bhey\b", r"\bhola\b", r"\bgreetings\b",
+        r"\bwhat('s| is) up\b", r"\bhowdy\b", r"\bhiya\b", r"\byo\b",
+        r"\bgood (morning|afternoon|evening|day|night)\b",
+        r"\bhow (are|r) you\b", r"\bhow's it going\b", r"\bhow have you been\b",
+        r"\bhope you are (doing )?(well|good|fine)\b", r"\bnice to meet you\b",
+        r"\bpleased to meet you\b"
+    ]
+
+    # Thank-you messages
+    thank_you_patterns = [
+        r"\bthank(s| you)\b", r"\bthanks a lot\b", r"\bthanks so much\b",
+        r"\bthank you very much\b", r"\bappreciate it\b", r"\bmuch obliged\b",
+        r"\bgrateful\b", r"\bcheers\b"
+    ]
+
+    # Small talk
+    small_talk_patterns = [
+        r"\bhow (are|r) you\b", r"\bhow's your day\b", r"\bwhat's up\b",
+        r"\bhow's it going\b", r"\bhow have you been\b", r"\bhope you are well\b"
+    ]
+
+    # Unrelated topics
+    unrelated_patterns = [
+        r"\btell me a joke\b", r"\bwho won\b", r"\bwhat is ai\b", r"\bexplain blockchain\b"
+    ]
+
+
+    # Goodbye messages
+    goodbye_patterns = [
+        r"\bbye\b", r"\bgoodbye\b", r"\bsee you\b", r"\bhave a nice day\b"
+    ]
+
+    # Rude or inappropriate messages
+    rude_patterns = [
+        r"\bstupid\b", r"\bdumb\b", r"\buseless\b", r"\bshut up\b"
+    ]
+
+    if any(re.search(pattern, query) for pattern in greeting_patterns):
+        return "greeting"
+    if any(re.search(pattern, query) for pattern in thank_you_patterns):
+        return "thank_you"
+    if any(re.search(pattern, query) for pattern in small_talk_patterns):
+        return "small_talk"
+    if any(re.search(pattern, query) for pattern in unrelated_patterns):
+        return "unrelated"
+    if any(re.search(pattern, query) for pattern in goodbye_patterns):
+        return "goodbye"
+    if any(re.search(pattern, query) for pattern in rude_patterns):
+        return "rude"
+
+    return "normal"
+
+# Function to Search for Relevant Answers
+def search_text(query, top_k=2):
+    query_embedding = np.array(model.encode(query, convert_to_numpy=True)).astype("float32").reshape(1, -1)
+    distances, indices = index.search(query_embedding, top_k)
+
+    results = []
+    for idx in indices[0]:
+        if idx >= 0:
+            results.append(metadata[idx])
+    
+    return results
+
+# Serve HTML Page
+@app.route("/")
+def home():
+    return render_template("index.html")
+
+@app.route("/query", methods=["POST"])
+def query_pdf():
+    query = request.json.get("query")
+
+    query_type = categorize_query(query)
+
+    if query_type == "greeting":
+        return jsonify({"text": "Hello! How can I assist you with Exelsys EasyHR?", "images": []})
+
+    if query_type == "thank_you":
+        return jsonify({"text": "You're welcome! How can I assist you further?", "images": []})
+
+    if query_type == "small_talk":
+        return jsonify({"text": "I'm here to assist with Exelsys EasyHR. How can I help?", "images": []})
+
+    if query_type == "unrelated":
+        return jsonify({"text": "I'm here to assist with Exelsys easyHR queries only.", "images": []})
+
+    if query_type == "vague":
+        return jsonify({"text": "Could you please provide more details?", "images": []})
+
+    if query_type == "goodbye":
+        return jsonify({"text": "You're welcome! Have a great day!", "images": []})
+
+    if query_type == "rude":
+        return jsonify({"text": "I'm here to assist you professionally.", "images": []})
+
+
+
+    # Search for relevant PDF content using FAISS
+    results = search_text(query, top_k=3)
+
+    if not results:
+        return jsonify({"text": "No relevant results found in the PDF.", "images": []})
+
+    # Merge multiple text results
+    retrieved_text = "\n\n---\n\n".join([res["text"] for res in results])
+    print(retrieved_text)
+    
+    prompt_extract = PromptTemplate.from_template(
+        """
+        ### YOU ARE AN EXELSYS EASYHR GUIDE ASSISTANT:
+        ### INSTRUCTIONS:
+        - Your job is to provide step-by-step guidance for the following user query based on the provided context.
+        - Base your response **only** on the retrieved context from the PDF.
+        - If no relevant information is found, simply respond with: "Not found."
+        - If the user greets you (e.g., "Hello", "Hi", "Good morning"), respond politely but keep it brief.
+        - If the query is unrelated to Exelsys easyHR, respond with: "I'm here to assist with Exelsys easyHR queries only."
+        - Provide clear and concise answers.
+        - Provide all the links that inside any topic in <a> tag.
+        
+
+        ### USER QUERY:
+        {query}
+
+        ### CONTEXT FROM PDF:
+        {retrieved_text}
+
+        ### ANSWER:
+        """
+    )
+
+    # Chain the prompt with ChatGroq
+    chain_extract = prompt_extract | llm
+    chat_response = chain_extract.invoke({"query": query, "retrieved_text": retrieved_text})
+
+    # Convert response to string
+    response_text = str(chat_response.content)
+
+    # Determine if images should be included
+    # retrieved_images = []
+    # if "Not found." not in response_text and "I'm here to assist" not in response_text:
+    #     retrieved_images = [img for res in results if "images" in res for img in res["images"]]
+
+    # Final response JSON
+    response = {
+        "text": response_text,
+        # "images": retrieved_images
+    }
+    print(response)
+
+    return jsonify(response)
+
+if __name__ == "__main__":
+    app.run(host="0.0.0.0", port=7860)
+