Upload 6 files

README.md

@@ -0,0 +1,39 @@
+# 🎓 University Inquiries AI Chatbot
+
+A conversational AI chatbot that helps students navigate and understand the official university handbook using natural language. Built with Gradio, Sentence Transformers, and FLAN-T5, it offers friendly and accurate responses based on handbook content — with built-in feedback tracking and upvote/downvote tuning.
+
+---
+
+## ✨ Features
+
+- 🧠 **Semantic Search:** Uses Sentence Transformers to find the closest handbook Q&A to a student's question.
+- 🗣️ **LLM Explanation:** Automatically rewrites formal handbook responses in a student-friendly tone using FLAN-T5.
+- 👍 **Feedback Memory:** Stores user feedback (upvotes/downvotes) to improve future responses.
+- 🔄 **Smart Matching:** Merges similar feedback questions (≥80% similarity) to avoid duplication.
+- 🎨 **Custom UI:** Includes a PUP-themed responsive design with a gradient background and styled chat layout.
+
+
+---
+
+## 🧰 Tech Stack
+
+- [Gradio](https://gradio.app/) — for UI interface
+- [Sentence Transformers](https://www.sbert.net/) — for question embedding & similarity
+- [Transformers (FLAN-T5)](https://huggingface.co/google/flan-t5-small) — for natural explanation generation
+- JSON — for persistent feedback storage
+
+---
+
+## 📁 Files
+
+- `app.py` — main chatbot logic & interface
+- `dataset.json` — official university Q&A set
+- `feedback.json` — live feedback database
+- `README.md` — this file!
+
+---
+
+## 📦 Installation
+
+```bash
+pip install -r requirements.txt

app.py

@@ -0,0 +1,183 @@
+import gradio as gr
+import json
+from sentence_transformers import SentenceTransformer
+from transformers import pipeline
+from sklearn.metrics.pairwise import cosine_similarity
+import numpy as np
+import os
+
+# === Custom PUP-themed CSS ===
+PUP_Themed_css = """
+html, body, .gradio-container, .gr-app {
+    height: 100% !important;
+    margin: 0 !important;
+    padding: 0 !important;
+    background: linear-gradient(to bottom right, #800000, #ff0000, #ffeb3b, #ffa500) !important;
+    font-family: 'Segoe UI', Tahoma, Geneva, Verdana, sans-serif !important;
+    color: #1b4332 !important;
+}
+"""
+
+# === Load Models and Data ===
+embedding_model = SentenceTransformer('paraphrase-mpnet-base-v2')
+llm = pipeline("text2text-generation", model="google/flan-t5-small")
+
+with open("dataset.json", "r") as f:
+    dataset = json.load(f)
+
+questions = [item["question"] for item in dataset]
+answers = [item["answer"] for item in dataset]
+question_embeddings = embedding_model.encode(questions, convert_to_tensor=True)
+
+chat_history = []
+feedback_data = []
+feedback_questions = []
+feedback_answers = []
+feedback_embeddings = None
+
+if os.path.exists("feedback.json") and os.path.getsize("feedback.json") > 0:
+    with open("feedback.json", "r") as f:
+        try:
+            feedback_data = json.load(f)
+            feedback_questions = [item["question"] for item in feedback_data]
+            feedback_answers = [item["response"] for item in feedback_data]
+            if feedback_questions:
+                feedback_embeddings = embedding_model.encode(feedback_questions, convert_to_tensor=True)
+        except json.JSONDecodeError:
+            feedback_data = []
+
+# === Chatbot Response Function ===
+def chatbot_response(query, chat_history):
+    query_embedding = embedding_model.encode([query], convert_to_tensor=True)
+
+    # === Feedback Matching ===
+    if feedback_embeddings is not None:
+        feedback_scores = cosine_similarity(query_embedding.cpu().numpy(), feedback_embeddings.cpu().numpy())[0]
+        best_idx = int(np.argmax(feedback_scores))
+        best_score = feedback_scores[best_idx]
+        matched_feedback = feedback_data[best_idx]
+
+        base_threshold = 0.8
+        upvotes = matched_feedback.get("upvotes", 0)
+        downvotes = matched_feedback.get("downvotes", 0)
+        adjusted_threshold = base_threshold - (0.01 * upvotes) + (0.01 * downvotes)
+        dynamic_threshold = min(max(adjusted_threshold, 0.4), 1.0)
+
+        if best_score >= dynamic_threshold:
+            response = matched_feedback["response"]
+            chat_history.append((query, response))
+            return "", chat_history, gr.update(visible=True)
+
+    # === Main Handbook Matching ===
+    similarity_scores = cosine_similarity(query_embedding.cpu().numpy(), question_embeddings.cpu().numpy())[0]
+    best_idx = int(np.argmax(similarity_scores))
+    best_score = similarity_scores[best_idx]
+    matched_q = questions[best_idx]
+    matched_a = answers[best_idx]
+
+    if best_score < 0.4:
+        response = "Sorry, I couldn't find a relevant answer."
+        chat_history.append((query, response))
+        return "", chat_history, gr.update(visible=True)
+
+    prompt = (
+        f"The following is an official university handbook statement:\n"
+        f"\"{matched_a}\"\n\n"
+        f"Please explain this to a student in a short, natural, and easy-to-understand way. "
+        f"Use simple words, and do not add new information."
+    )
+
+    llm_response = llm(prompt, max_length=200, do_sample=True, temperature=0.7, top_p=0.9)[0]["generated_text"].strip()
+    if not llm_response:
+        llm_response = "I'm sorry, I couldn't simplify that at the moment."
+
+    a_embedding = embedding_model.encode([matched_a], convert_to_tensor=True)
+    llm_embedding = embedding_model.encode([llm_response], convert_to_tensor=True)
+    explanation_similarity = cosine_similarity(a_embedding.cpu().numpy(), llm_embedding.cpu().numpy())[0][0]
+
+    if explanation_similarity >= 0.95:
+        final_response = f"According to the university handbook, {matched_a}"
+    else:
+        final_response = f"According to the university handbook, {matched_a} In simpler terms, {llm_response}"
+
+    chat_history.append((query, final_response))
+    return "", chat_history, gr.update(visible=True)
+
+# === Feedback Save & Upvote/Downvote Tracking ===
+def record_feedback(feedback, chat_history):
+    global feedback_embeddings
+    if chat_history:
+        last_query, last_response = chat_history[-1]
+        matched = False
+
+        for item in feedback_data:
+            existing_embedding = embedding_model.encode([item["question"]], convert_to_tensor=True)
+            new_embedding = embedding_model.encode([last_query], convert_to_tensor=True)
+            similarity = cosine_similarity(existing_embedding.cpu().numpy(), new_embedding.cpu().numpy())[0][0]
+            if similarity >= 0.8 and item["response"] == last_response:
+                matched = True
+                votes = {"positive": "upvotes", "negative": "downvotes"}
+                item[votes[feedback]] = item.get(votes[feedback], 0) + 1
+                break
+
+        if not matched:
+            entry = {
+                "question": last_query,
+                "response": last_response,
+                "feedback": feedback,
+                "upvotes": 1 if feedback == "positive" else 0,
+                "downvotes": 1 if feedback == "negative" else 0
+            }
+            feedback_data.append(entry)
+
+        with open("feedback.json", "w") as f:
+            json.dump(feedback_data, f, indent=4)
+
+        # Update feedback embeddings
+        feedback_questions = [item["question"] for item in feedback_data]
+        if feedback_questions:
+            feedback_embeddings = embedding_model.encode(feedback_questions, convert_to_tensor=True)
+
+    return gr.update(visible=False)
+
+# === Gradio UI ===
+with gr.Blocks(css=PUP_Themed_css, title="University Handbook AI Chatbot") as demo:
+    gr.Markdown(
+        "<div style='"
+        "background-color: #ffffff; "
+        "border-radius: 16px; "
+        "padding: 24px 16px; "
+        "margin-bottom: 24px; "
+        "box-shadow: 0 6px 16px rgba(0, 0, 0, 0.15); "
+        "max-width: 700px; "
+        "margin-left: auto; "
+        "margin-right: auto; "
+        "text-align: center;'>"
+        "<h1 style='font-size: 2.2rem; margin: 0;'>University Inquiries AI Chatbot</h1>"
+        "</div>"
+    )
+
+    state = gr.State(chat_history)
+    chatbot_ui = gr.Chatbot(label="Chat", show_label=False)
+
+    with gr.Row():
+        query_input = gr.Textbox(placeholder="Type your question here...", show_label=False)
+        submit_btn = gr.Button("Submit")
+
+    with gr.Row(visible=False) as feedback_row:
+        gr.Markdown("Was this helpful?")
+        thumbs_up = gr.Button("👍")
+        thumbs_down = gr.Button("👎")
+
+    def handle_submit(message, chat_state):
+        return chatbot_response(message, chat_state)
+
+    submit_btn.click(handle_submit, [query_input, state], [query_input, chatbot_ui, feedback_row])
+    query_input.submit(handle_submit, [query_input, state], [query_input, chatbot_ui, feedback_row])
+
+    thumbs_up.click(lambda state: record_feedback("positive", state), inputs=[state], outputs=[feedback_row])
+    thumbs_down.click(lambda state: record_feedback("negative", state), inputs=[state], outputs=[feedback_row])
+
+# === Launch App ===
+if __name__ == "__main__":
+    demo.launch()

feedback.json

@@ -0,0 +1 @@
+[]

requirements.txt

@@ -0,0 +1,5 @@
+gradio
+transformers
+sentence-transformers
+scikit-learn
+numpy

space.yaml

@@ -0,0 +1,8 @@
+# space.yaml
+title: "University Inquiries AI Chatbot"
+emoji: 🎓
+colorFrom: red
+colorTo: yellow
+sdk: gradio
+python_version: 3.10
+app_file: app.py