Upload 4 files

app.py

@@ -0,0 +1,33 @@
+import gradio as gr
+import json
+from utils import get_best_answer
+from dotenv import load_dotenv
+
+# ✅ Load environment variables
+load_dotenv()
+
+# ✅ Load dataset
+with open("dataset.json", "r") as f:
+    dataset = json.load(f)
+
+# ✅ Gradio UI Function
+def chatbot_response(user_input):
+    try:
+        best_answer = get_best_answer(user_input)
+        return best_answer  # ✅ Displays the rephrased best answer
+    except Exception as e:
+        return "Oops! Something went wrong. Please try again."
+
+# ✅ Gradio Interface
+iface = gr.Interface(
+    fn=chatbot_response,
+    inputs="text",
+    outputs="text",
+    title="UOE Academic Chatbot",
+    description="Hello! 🎓 Welcome to **Academic Navigator** – the university information assistant for UOE.\nI can answer your university-related questions, such as admissions, scholarships."
+)
+
+# ✅ Launch App
+iface.launch(share=True)
+
+

requirements.txt

@@ -0,0 +1,7 @@
+gradio
+sentence-transformers
+torch
+transformers
+requests
+python-dotenv
+      # ✅ For spelling correction using TextBlob

utils.py

@@ -0,0 +1,89 @@
+import os
+import json
+import requests
+from sentence_transformers import SentenceTransformer, util
+from dotenv import load_dotenv
+
+
+# ✅ Load .env file
+load_dotenv()
+
+# ✅ Hugging Face API Token check
+HF_API_TOKEN = os.getenv("HF_API_TOKEN")
+
+if not HF_API_TOKEN:
+    raise ValueError("Error: Hugging Face API Token is missing! Please check your .env file.")
+
+# ✅ Hugging Face GPT Model API Endpoint
+GPT_MODEL_API = "https://api-inference.huggingface.co/models/mistralai/Mistral-7B-Instruct-v0.2"
+
+# ✅ Headers for API request
+headers = {"Authorization": f"Bearer {HF_API_TOKEN}"}
+
+# ✅ Load sentence transformer model for intent matching
+similarity_model = SentenceTransformer('paraphrase-MiniLM-L6-v2')
+
+# ✅ Load dataset
+with open("dataset.json") as f:
+    dataset = json.load(f)
+
+# ✅ Precompute embeddings for dataset questions
+dataset_inputs = [item.get("input", "").lower().strip() for item in dataset]
+dataset_answers = [item.get("response", "") for item in dataset]
+dataset_embeddings = similarity_model.encode(dataset_inputs, convert_to_tensor=True)
+
+# ✅ Function to detect greetings
+def is_greeting(text):
+    greetings = ["hello", "hi", "hey", "salam", "assalam o alaikum", "assalamu alaikum", "assalamualaikum"]
+    return text.lower().strip() in greetings
+
+# ✅ Function to get the best matching answer
+def get_best_answer(user_input):
+    if is_greeting(user_input):
+        return "Hello! 😊 How can I assist you today with university-related information?"
+
+    user_input_embedding = similarity_model.encode(user_input.lower().strip(), convert_to_tensor=True)
+    similarities = util.pytorch_cos_sim(user_input_embedding, dataset_embeddings)[0]
+
+    # ✅ Find the best match and its similarity score
+    best_match_index = similarities.argmax().item()
+    best_score = similarities[best_match_index].item()
+
+    # ✅ Set a similarity threshold (tune as needed)
+    THRESHOLD = 0.65
+
+    if best_score < THRESHOLD:
+        return (
+             "I'm sorry, I couldn't find an exact answer to your question. "
+             "You may kindly try rephrasing your question gently for better results. "
+             "Also, feel free to visit the UOE official website for information: https://ue.edu.pk/"
+        )
+
+    best_answer = dataset_answers[best_match_index]
+    return rephrase_answer(best_answer)
+
+# ✅ Function to rephrase answer using GPT
+def rephrase_answer(answer):
+    prompt = (
+        f"Rephrase the following university-related answer while keeping the meaning unchanged:\n\n"
+        f"Original Answer: {answer}\n\n"
+        f"Rephrased Answer:"
+    )
+
+    payload = {"inputs": prompt}  # ✅ Structured prompt to prevent extra details
+
+    response = requests.post(GPT_MODEL_API, headers=headers, json=payload)
+
+    if response.status_code == 200:
+        result = response.json()
+        if isinstance(result, list) and result:
+            generated_text = result[0].get("generated_text", answer).strip()
+            
+            # ✅ Ensure only the rephrased answer is returned
+            if "Rephrased Answer:" in generated_text:
+                return generated_text.split("Rephrased Answer:")[-1].strip()
+            return generated_text
+
+    return answer  # ✅ Return original answer if API fails
+
+