Update app.py

app.py

@@ -4,6 +4,9 @@ from transformers import AutoTokenizer, AutoModelForCausalLM
 from duckduckgo_search import DDGS
 import time
 import torch
+import base64
+import tempfile
+import os
 
 # Initialize models
 print("Loading Whisper model...")
@@ -23,7 +26,7 @@ model = AutoModelForCausalLM.from_pretrained(
 ddgs = DDGS(timeout=3)
 
 def search_web(query, max_results=2):
-    """Perform web search using DuckDuckGo (FREE & UNLIMITED)"""
+    """Perform web search using DuckDuckGo"""
     try:
         results = ddgs.text(
             keywords=query,
@@ -44,8 +47,70 @@ def search_web(query, max_results=2):
     except Exception as e:
         return f"Search failed: {str(e)}"
 
+def transcribe_audio_base64(audio_base64):
+    """Transcribe audio from base64 string (for Pluely STT endpoint)"""
+    try:
+        # Decode base64 audio
+        audio_bytes = base64.b64decode(audio_base64)
+        
+        # Save to temporary file
+        with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as temp_audio:
+            temp_audio.write(audio_bytes)
+            temp_path = temp_audio.name
+        
+        # Transcribe
+        segments, _ = whisper_model.transcribe(temp_path, language="en", beam_size=1)
+        transcription = " ".join([seg.text for seg in segments])
+        
+        # Cleanup
+        os.unlink(temp_path)
+        
+        return {"text": transcription.strip()}
+    
+    except Exception as e:
+        return {"error": f"Transcription failed: {str(e)}"}
+
+def generate_answer(text_input):
+    """Generate answer from text input (for Pluely AI endpoint)"""
+    try:
+        if not text_input or text_input.strip() == "":
+            return "No input provided"
+        
+        # Web search
+        search_results = search_web(text_input, max_results=2)
+        
+        # Generate answer
+        messages = [
+            {"role": "system", "content": "You are a helpful assistant. Answer briefly using provided context. Keep responses under 40 words."},
+            {"role": "user", "content": f"Context:\n{search_results}\n\nQuestion: {text_input}\n\nAnswer:"}
+        ]
+        
+        text = tokenizer.apply_chat_template(
+            messages,
+            tokenize=False,
+            add_generation_prompt=True
+        )
+        
+        inputs = tokenizer([text], return_tensors="pt").to("cpu")
+        
+        with torch.no_grad():
+            outputs = model.generate(
+                **inputs,
+                max_new_tokens=80,
+                temperature=0.2,
+                do_sample=True,
+                top_p=0.85,
+                pad_token_id=tokenizer.eos_token_id
+            )
+        
+        response = tokenizer.decode(outputs[0][inputs['input_ids'].shape[1]:], skip_special_tokens=True)
+        return response.strip()
+        
+    except Exception as e:
+        return f"Error: {str(e)}"
+
 def process_audio(audio_path, question_text=None):
-    """Main pipeline: audio -> text -> search -> answer"""
+    """Main pipeline for Gradio UI"""
     start_time = time.time()
     
     # Step 1: Transcribe audio if provided
@@ -68,52 +133,22 @@ def process_audio(audio_path, question_text=None):
     search_results = search_web(question, max_results=2)
     search_time = time.time() - search_start
     
-    # Step 3: Generate answer with LLM
+    # Step 3: Generate answer
     llm_start = time.time()
-    
-    messages = [
-        {"role": "system", "content": "You are a helpful assistant. Answer questions briefly using the provided context."},
-        {"role": "user", "content": f"Context:\n{search_results}\n\nQuestion: {question}\n\nAnswer:"}
-    ]
-    
-    try:
-        text = tokenizer.apply_chat_template(
-            messages,
-            tokenize=False,
-            add_generation_prompt=True
-        )
-        
-        inputs = tokenizer([text], return_tensors="pt").to("cpu")
-        
-        with torch.no_grad():
-            outputs = model.generate(
-                **inputs,
-                max_new_tokens=120,
-                temperature=0.2,
-                do_sample=True,
-                top_p=0.85,
-                pad_token_id=tokenizer.eos_token_id
-            )
-        
-        response = tokenizer.decode(outputs[0][inputs['input_ids'].shape[1]:], skip_special_tokens=True)
-        answer = response.strip()
-        
-    except Exception as e:
-        answer = f"❌ LLM error: {str(e)}"
-    
+    answer = generate_answer(question)
     llm_time = time.time() - llm_start
-    total_time = time.time() - start_time
     
+    total_time = time.time() - start_time
     time_emoji = "🟢" if total_time < 3.0 else "🟡" if total_time < 3.5 else "🔴"
     timing_info = f"\n\n{time_emoji} **Timing:** Trans={transcription_time:.2f}s | Search={search_time:.2f}s | LLM={llm_time:.2f}s | **Total={total_time:.2f}s**"
     
     return answer + timing_info, total_time
 
-# Create Gradio interface (same as before)
-with gr.Blocks(title="Fast Q&A - No Building Required!", theme=gr.themes.Soft()) as demo:
+# Create Gradio interface with API endpoints
+with gr.Blocks(title="Fast Q&A - Pluely Compatible", theme=gr.themes.Soft()) as demo:
     gr.Markdown("""
     # ⚡ Ultra-Fast Political Q&A System
-    **No wheel building** - Fast deployment with transformers!
+    **Pluely Compatible** - Direct STT and AI endpoints available!
     
     **Features:** Whisper-tiny + Qwen2.5-0.5B + DuckDuckGo (FREE unlimited search)
     """)
@@ -169,11 +204,72 @@ with gr.Blocks(title="Fast Q&A - No Building Required!", theme=gr.themes.Soft())
             inputs=text_input
         )
     
+    # Hidden API endpoints for Pluely
+    with gr.Tab("🔌 Pluely Integration"):
+        gr.Markdown("""
+        ## Dedicated Endpoints for Pluely
+        
+        ### 1. STT Endpoint (Audio Transcription)
+        ```
+        curl -X POST https://archcoder-basic-app.hf.space/call/transcribe_stt \\
+          -H "Content-Type: application/json" \\
+          -d '{"data": ["BASE64_AUDIO_DATA"]}'
+        ```
+        **Returns:** `{"data": [{"text": "transcribed text"}]}`
+        
+        ### 2. AI Endpoint (Text to Answer)
+        ```
+        curl -X POST https://archcoder-basic-app.hf.space/call/answer_ai \\
+          -H "Content-Type: application/json" \\
+          -d '{"data": ["Your question here"]}'
+        ```
+        **Returns:** `{"data": ["Answer text"]}`
+        
+        ---
+        
+        ## Pluely Configuration
+        
+        ### Custom STT Provider:
+        **Curl Command:**
+        ```
+        curl --location 'https://archcoder-basic-app.hf.space/call/transcribe_stt' \\
+          --header 'Content-Type: application/json' \\
+          --data '{"data": ["{{AUDIO_BASE64}}"]}'
+        ```
+        **Response Content Path:** `data[0].text`
+        
+        ### Custom AI Provider:
+        **Curl Command:**
+        ```
+        curl --location 'https://archcoder-basic-app.hf.space/call/answer_ai' \\
+          --header 'Content-Type: application/json' \\
+          --data '{"data": ["{{TEXT}}"]}'
+        ```
+        **Response Content Path:** `data[0]`
+        """)
+    
     gr.Markdown("""
     ---
     🟢 = Under 3s | 🟡 = 3-3.5s | 🔴 = Over 3.5s
     """)
 
+# Register API endpoints
+demo.api_name = "pluely_integration"
+
+# STT endpoint for Pluely
+@demo.api(api_name="transcribe_stt")
+def api_transcribe(audio_base64: str):
+    """API endpoint for audio transcription (Pluely STT)"""
+    result = transcribe_audio_base64(audio_base64)
+    return result
+
+# AI endpoint for Pluely
+@demo.api(api_name="answer_ai")
+def api_answer(text: str):
+    """API endpoint for text-to-answer (Pluely AI)"""
+    answer = generate_answer(text)
+    return answer
+
 if __name__ == "__main__":
     demo.queue(max_size=5)
     demo.launch()