app.py

"""
AI Graduate Job-Matching Platform
Production-Safe Version with Retry & Chunk Handling
Developer: Najaf Ali Sharqi
MoviePy-free (uses ffmpeg + tempfile)
"""

import os
import gradio as gr
import tempfile
import subprocess
import math
import time
from groq import Groq

# ============================================================================
# CONFIGURATION
# ============================================================================
client = Groq(api_key=os.getenv("GROQ_API_KEY"))

TRANSCRIPTION_MODEL = "whisper-large-v3"
REWRITE_MODEL = "openai/gpt-oss-120b"
CHUNK_SIZE = 300  # 5 min chunks for transcription to avoid rate limit
REWRITE_CHUNK_SIZE = 4000  # words per rewrite chunk
MAX_RETRIES = 5  # retries for 429 errors

# ============================================================================
# UTILITY FUNCTIONS
# ============================================================================
def get_video_duration(video_path):
    """Get video duration using ffprobe"""
    try:
        result = subprocess.run(
            ["ffprobe", "-v", "error", "-show_entries",
             "format=duration", "-of",
             "default=noprint_wrappers=1:nokey=1", video_path],
            stdout=subprocess.PIPE,
            stderr=subprocess.STDOUT,
            text=True
        )
        duration = float(result.stdout.strip())
        return duration
    except Exception as e:
        raise RuntimeError(f"Failed to get video duration: {str(e)}")

def extract_audio_chunks(video_path, chunk_size=CHUNK_SIZE):
    """Extract audio from video and split into chunks"""
    try:
        duration = get_video_duration(video_path)
        num_chunks = math.ceil(duration / chunk_size)
        audio_chunks = []

        # extract full audio first
        temp_audio_full = tempfile.NamedTemporaryFile(suffix=".mp3", delete=False)
        audio_full_path = temp_audio_full.name
        temp_audio_full.close()

        os.system(f"ffmpeg -y -i \"{video_path}\" -q:a 0 -map a \"{audio_full_path}\"")

        # split audio into chunks
        for i in range(num_chunks):
            start = i * chunk_size
            end = min((i + 1) * chunk_size, duration)
            temp_chunk = tempfile.NamedTemporaryFile(suffix=".mp3", delete=False)
            chunk_path = temp_chunk.name
            temp_chunk.close()
            os.system(f"ffmpeg -y -i \"{audio_full_path}\" -ss {start} -to {end} -c copy \"{chunk_path}\"")
            audio_chunks.append(chunk_path)

        # remove full audio file
        if os.path.exists(audio_full_path):
            os.unlink(audio_full_path)

        return audio_chunks
    except Exception as e:
        raise RuntimeError(f"Audio extraction failed: {str(e)}")

# ============================================================================
# TRANSCRIPTION WITH RETRY
# ============================================================================
def transcribe_audio_chunks(audio_chunks, language="English"):
    transcript_text = ""
    for chunk in audio_chunks:
        for attempt in range(MAX_RETRIES):
            try:
                with open(chunk, "rb") as audio_file:
                    transcription = client.audio.transcriptions.create(
                        file=(os.path.basename(chunk), audio_file.read()),
                        model=TRANSCRIPTION_MODEL,
                        temperature=0,
                        response_format="verbose_json",
                        language="en" if language=="English" else "ur"
                    )
                transcript_text += transcription.text + "\n\n"
                break  # success
            except Exception as e:
                err_str = str(e)
                if "Rate limit reached" in err_str or "429" in err_str:
                    wait_time = 5 + attempt * 5
                    print(f"Rate limit hit. Waiting {wait_time}s before retrying...")
                    time.sleep(wait_time)
                else:
                    transcript_text += f"[Error in chunk transcription: {err_str}]\n\n"
                    break
        if os.path.exists(chunk):
            os.unlink(chunk)
    return transcript_text.strip()

# ============================================================================
# GRAMMAR & PUNCTUATION CORRECTION (Chunked)
# ============================================================================
def rewrite_transcript_chunked(text, language="English", chunk_size=REWRITE_CHUNK_SIZE):
    """Rewrite transcript in smaller chunks to avoid token limits"""
    try:
        words = text.split()
        rewritten_chunks = []
        for i in range(0, len(words), chunk_size):
            chunk_text = " ".join(words[i:i+chunk_size])
            prompt = f"Correct grammar, punctuation, and make it readable. Language: {language}\n\n{chunk_text}"
            for attempt in range(MAX_RETRIES):
                try:
                    response = client.chat.completions.create(
                        model=REWRITE_MODEL,
                        messages=[{"role": "user", "content": prompt}],
                        temperature=0
                    )
                    rewritten_chunks.append(response.choices[0].message.content)
                    break
                except Exception as e:
                    err_str = str(e)
                    if "Rate limit reached" in err_str or "429" in err_str:
                        wait_time = 5 + attempt*5
                        print(f"Rewrite rate limit hit. Waiting {wait_time}s before retrying...")
                        time.sleep(wait_time)
                    else:
                        rewritten_chunks.append(f"[Error rewriting chunk: {err_str}]")
                        break
        return "\n\n".join(rewritten_chunks)
    except Exception as e:
        return f"[Error rewriting transcript: {str(e)}]"

def correct_transcript(transcript_text, language):
    lang = "English" if "English" in language else "Urdu"
    if not transcript_text:
        return "", "Please generate transcript first." if lang=="English" else "پہلے ٹرانسکرپٹ بنائیں۔"
    rewritten_text = rewrite_transcript_chunked(transcript_text, language=lang)
    return rewritten_text, "✅ Grammar & punctuation corrected / گرامر اور پنکچویشن درست کی گئی"

# ============================================================================
# PROCESS VIDEO
# ============================================================================
def generate_transcript(video_file, language):
    lang = "English" if "English" in language else "Urdu"
    if not video_file:
        return "", "Please upload a video file." if lang=="English" else "براہ کرم ایک ویڈیو فائل اپ لوڈ کریں۔"
    try:
        audio_chunks = extract_audio_chunks(video_file.name)
        transcript = transcribe_audio_chunks(audio_chunks, language=lang)
        return transcript, "✅ Transcription completed / ٹرانسکرپشن مکمل"
    except Exception as e:
        error_msg = str(e)
        return "", error_msg if lang=="English" else "خرابی: " + error_msg

# ============================================================================
# DOWNLOAD FUNCTION
# ============================================================================
def download_transcript(text, language):
    lang_suffix = "en" if "English" in language else "ur"
    temp_file = tempfile.NamedTemporaryFile(delete=False, suffix=f"_{lang_suffix}.txt")
    temp_file.write(text.encode("utf-8"))
    temp_file.close()
    return temp_file.name

# ============================================================================
# GRADIO UI
# ============================================================================
def create_ui():
    with gr.Blocks(title="AI Graduate Job Matcher") as demo:
        gr.HTML("<h1>🎓 AI Graduate Job-Matching Platform</h1><p>Developer: Najaf Ali Sharqi</p>")
        
        video_input = gr.File(label="Upload Video / ویڈیو اپ لوڈ کریں", file_types=[".mp4", ".mov", ".avi"])
        language_dropdown = gr.Dropdown(choices=["English", "Urdu / اردو"], value="English", label="Output Language / آؤٹ پٹ زبان")
        
        transcript_output = gr.Textbox(label="Transcript / ٹرانسکرپٹ", lines=20)
        status_output = gr.Textbox(label="Status / حیثیت", lines=1)
        
        # Buttons
        transcribe_btn = gr.Button("🚀 Generate Transcript / ٹرانسکرپٹ بنائیں")
        rewrite_btn = gr.Button("✏️ Correct Grammar & Punctuation / گرامر درست کریں")
        download_btn = gr.Button("💾 Download Transcript / ٹرانسکرپٹ ڈاؤن لوڈ کریں")
        
        # Transcription click
        transcribe_btn.click(
            fn=generate_transcript,
            inputs=[video_input, language_dropdown],
            outputs=[transcript_output, status_output]
        )
        
        # Grammar correction click
        rewrite_btn.click(
            fn=correct_transcript,
            inputs=[transcript_output, language_dropdown],
            outputs=[transcript_output, status_output]
        )
        
        # Download click
        download_btn.click(
            fn=download_transcript,
            inputs=[transcript_output, language_dropdown],
            outputs=gr.File(label="Download your transcript")
        )
    
    return demo

# ============================================================================
# MAIN
# ============================================================================
if __name__ == "__main__":
    if not os.getenv("GROQ_API_KEY"):
        print("⚠️ WARNING: GROQ_API_KEY not found!")
    demo = create_ui()
    demo.launch(server_name="0.0.0.0", server_port=7860)