Upload 2 files

app.py

@@ -0,0 +1,211 @@
+import gradio as gr
+import pyttsx3
+import PyPDF2
+import os
+import time
+import uuid
+import numpy as np
+from gtts import gTTS
+from playsound import playsound
+from sentence_transformers import SentenceTransformer
+import chromadb
+from groq import Groq
+
+
+import os
+from dotenv import load_dotenv
+load_dotenv()
+GROQ_API_KEY = os.environ.get("GROQ_API_KEY")
+
+groq_client = Groq(api_key=GROQ_API_KEY)
+model = SentenceTransformer('all-MiniLM-L6-v2')
+client = chromadb.Client()
+collection = client.create_collection("echo_study")
+PDF_FOLDER = "pdfs"
+loaded_files = {}
+pdf_texts = {}
+current_file = {"name": None}
+
+
+QUESTIONS = {
+    "embedded systems": [
+        "How does the lecture define an Embedded System?",
+        "What are the primary resource constraints in embedded systems?",
+        "How do embedded systems interact with the physical world?"
+    ],
+    "dynamic programming": [
+        "What is the simplest way to define Dynamic Programming?",
+        "How many times does DP solve each subproblem?",
+        "What is the simple formula for Dynamic Programming"
+    ],
+    "mongol history": [
+        "Why did the Empire's huge size lead to its fall?",
+        "What was the original goal of the British East India Company?"
+    ]
+}
+
+
+def speak_system(text):
+    engine = pyttsx3.init()
+    engine.setProperty('rate', 140)
+    engine.say(text)
+    engine.runAndWait()
+
+def speak_user(text):
+    audio_path = f"C:/Users/hnaal/Desktop/Echo_study/user_{uuid.uuid4()}.mp3"
+    tts = gTTS(text=text, lang='en')
+    tts.save(audio_path)
+    playsound(audio_path)
+    os.remove(audio_path)
+
+
+def load_all_pdfs():
+    speak_system("Welcome back! Ready to tackle your studies?")
+    yield "⏳ Processing Embeddings..."
+    for filename in os.listdir(PDF_FOLDER):
+        if filename.endswith(".pdf"):
+            filepath = os.path.join(PDF_FOLDER, filename)
+            with open(filepath, "rb") as f:
+                reader = PyPDF2.PdfReader(f)
+                text = ""
+                for page in reader.pages:
+                    text += page.extract_text()
+                pdf_texts[filename] = text
+                embedding = model.encode(text[:2000]).tolist()
+                collection.add(
+                    documents=[text[:2000]],
+                    embeddings=[embedding],
+                    ids=[filename],
+                    metadatas=[{"source": filename}]
+                )
+                name = filename.replace(".pdf", "").replace("_", " ").lower()
+                loaded_files[name] = filename
+                yield f"⏳ Processing: {filename}..."
+    speak_system("All files loaded successfully.")
+    yield "✅ Loaded: " + ", ".join(loaded_files.keys())
+
+
+
+def update_questions(pdf_name):
+    pdf_key = pdf_name.lower()
+    for key in QUESTIONS:
+        if any(word in pdf_key for word in key.split()):
+            return gr.Dropdown(choices=QUESTIONS[key], value=QUESTIONS[key][0])
+    return gr.Dropdown(choices=[], value=None)
+
+def find_best_chunk(question, pdf_text):
+    chunks = []
+    words = pdf_text.split()
+    for i in range(0, len(words), 80):
+        chunk = " ".join(words[i:i+80])
+        chunks.append(chunk)
+    if not chunks:
+        return pdf_text[:500]
+    question_embedding = model.encode(question)
+    chunk_embeddings = [model.encode(chunk) for chunk in chunks]
+    similarities = [
+        np.dot(question_embedding, ce) / (np.linalg.norm(question_embedding) * np.linalg.norm(ce))
+        for ce in chunk_embeddings
+    ]
+    best_idx = similarities.index(max(similarities))
+    return chunks[best_idx]
+
+
+def ask_groq(question, context, file_name):
+    response = groq_client.chat.completions.create(
+        model="llama-3.3-70b-versatile",
+        messages=[
+            {
+                "role": "system",
+                "content": f"""You are EchoStudy, a warm and encouraging study partner for blind students.
+When answering:
+1. Start with a different warm phrase each time, like: Great question!, Interesting!, Good thinking!, Let me help you with that!
+2. Use a simple real-life analogy to explain if needed.
+3. Answer in 2 short sentences only, very simple and brief.
+4. Avoid markdown symbols like stars or hashtags.
+5. End with: Would you like more details?"""
+            },
+            {
+                "role": "user",
+                "content": f"Context from {file_name}: {context}\n\nQuestion: {question}"
+            }
+        ],
+        max_tokens=80
+    )
+    return response.choices[0].message.content
+
+
+def demo_interaction(pdf_name, question):
+    log = ""
+    if pdf_name.strip().lower() != current_file["name"]:
+        speak_system("Please say the name of your PDF file.")
+        log += "🔊 System: Please say the name of your PDF file.\n"
+        yield log, ""
+        time.sleep(1)
+        speak_user(pdf_name)
+        log += f"🎤 Student: {pdf_name}\n"
+        yield log, ""
+        time.sleep(1)
+        found = None
+        for name in loaded_files:
+            if any(word.lower() in name.lower() for word in pdf_name.split()):
+                found = name
+                break
+        if not found:
+            speak_system("Sorry, I could not find that file.")
+            log += "🔊 System: Sorry, I could not find that file.\n"
+            yield log, "Not found"
+            return
+        current_file["name"] = pdf_name.strip().lower()
+        speak_system(f"Found {found}. Ready for your question.")
+        log += f"🔊 System: Found {found}. Ready for your question.\n"
+        yield log, found
+        time.sleep(1)
+    else:
+        found = None
+        for name in loaded_files:
+            if any(word.lower() in name.lower() for word in pdf_name.split()):
+                found = name
+                break
+        if not found:
+            yield "File not found.", "Not found"
+            return
+    speak_user(question)
+    log += f"🎤 Student: {question}\n"
+    yield log, found
+    time.sleep(1)
+    target_file = loaded_files[found]
+    pdf_text = pdf_texts[target_file]
+    context = find_best_chunk(question, pdf_text)
+    answer = ask_groq(question, context, found)
+    speak_system(answer)
+    log += f"🔊 System: {answer}\n"
+    yield log, found
+
+
+
+with gr.Blocks() as app:
+    gr.Markdown("# 🎓 Echo Study – Voice-First Study Assistant")
+    gr.Markdown("♿ Designed for visually impaired students")
+    with gr.Row():
+        load_btn = gr.Button("📂 Load Study Materials")
+        load_status = gr.Textbox(label="Status")
+    gr.Markdown("### 🎤 Demo Interaction")
+    pdf_input = gr.Dropdown(
+        choices=["embedded systems", "dynamic programming", "mongol history"],
+        value="embedded systems",
+        label="📄 Select PDF"
+    )
+    question_input = gr.Dropdown(
+        choices=QUESTIONS["embedded systems"],
+        value=QUESTIONS["embedded systems"][0],
+        label="❓ Select Question"
+    )
+    selected_file = gr.Textbox(label="📂 Selected File", interactive=False)
+    start_btn = gr.Button("▶️ Start Demo", variant="primary")
+    conversation_log = gr.Textbox(label="Conversation Log", lines=10)
+    pdf_input.change(update_questions, inputs=pdf_input, outputs=question_input)
+    load_btn.click(load_all_pdfs, outputs=load_status, show_progress=False)
+    start_btn.click(demo_interaction, inputs=[pdf_input, question_input], outputs=[conversation_log, selected_file])
+
+app.launch()

requirements.txt

@@ -0,0 +1,10 @@
+gradio
+pyttsx3
+PyPDF2
+sentence-transformers
+chromadb
+gtts
+playsound==1.2.2
+groq
+python-dotenv
+numpy