Update server.py

server.py

@@ -149,36 +149,53 @@ async def process_transcription_for_highlights(
         )
         
         # Create structured prompt for segment extraction
-        system_prompt = """You are a movie marketing expert who identifies the most engaging and thrilling segments of movies.
-You will receive a full movie transcript with timestamps. Your task is to identify exactly 10 of the most compelling moments that would make audiences want to watch the full movie.
+        system_prompt = """You are an expert film editor and marketing strategist who identifies KEY MOMENTS from movies that make viewers WANT to watch the entire film.
 
-IMPORTANT: 
-- You MUST respond with a valid JSON array ONLY. Do not include any text before or after the JSON array.
-- Each segment should be approximately 10 MINUTES LONG (600 seconds)
-- End time = Start time + ~10 minutes
+Your task: Analyze the full movie transcript and identify exactly 10 HIGH-IMPACT moments that would serve as compelling hooks to attract audiences. These should be:
+- Plot turning points / major story beats
+- Action sequences / high tension moments
+- Emotional climaxes / character breakthroughs
+- Shocking reveals / plot twists
+- Character introductions / pivotal interactions
+- Climactic confrontations or resolutions
 
-Each segment must have:
-- segment_number: (1-10)
-- title: (engaging, compelling title for this moment)
-- start_time: (HH:MM:SS format - when this segment starts)
-- end_time: (HH:MM:SS format - when this segment ends, approximately 10 minutes after start)
-- description: (brief description of why this is engaging)
-- engagement_level: (high/medium)
-- reason: (one-line reason this will hook viewers)
+CRITICAL REQUIREMENTS:
+- You MUST respond with ONLY a valid JSON array. NO text before or after.
+- DO NOT arrange segments by time order - arrange them by ENGAGEMENT and IMPACT level
+- Segments can start ANYWHERE in the movie (5 min, 40 min, 90 min - doesn't matter)
+- Each segment should be 8-15 MINUTES LONG to capture the complete moment with context
+- Include exact timestamps from the transcript
+- Prioritize QUALITY over chronology - pick genuinely attention-grabbing moments
+- Segment 1 = MOST IMPACTFUL, Segment 10 = still engaging but slightly less
 
-Return ONLY the JSON array. Example format:
+Each segment must include:
+- segment_number: (1-10, ranked by engagement level, NOT by time)
+- title: (compelling, hooks viewers immediately)
+- start_time: (HH:MM:SS - exact start from transcript, can be anywhere)
+- end_time: (HH:MM:SS - exact end from transcript, captures full moment)
+- description: (2-3 sentences explaining why this moment is essential)
+- engagement_level: (high/medium - high for critical moments)
+- reason: (one line hook - what makes viewers want MORE)
+
+JSON format:
 [
-    {"segment_number": 1, "title": "Epic Action Scene", "start_time": "00:15:30", "end_time": "00:25:30", "description": "...", "engagement_level": "high", "reason": "..."},
-    {"segment_number": 2, "title": "Emotional Climax", "start_time": "00:45:12", "end_time": "00:55:12", "description": "...", "engagement_level": "high", "reason": "..."}
+    {"segment_number": 1, "title": "...", "start_time": "HH:MM:SS", "end_time": "HH:MM:SS", "description": "...", "engagement_level": "high", "reason": "..."}
 ]
 """
         
-        user_message = f"""Please extract exactly 10 of the most engaging segments from this movie transcript.
+        user_message = f"""Analyze this COMPLETE movie transcript and extract exactly 10 KEY MOMENTS that would make someone want to watch the full film.
+
+IMPORTANT: Do NOT arrange segments by when they appear in the movie. Arrange them by IMPACT and ENGAGEMENT.
+- A key scene at 40 minutes into the movie can be segment 1 if it's the most engaging
+- A scene at 10 minutes can be segment 8 if it's less impactful
+- Order by viewer attention level, NOT by timeline
+
+Find the 10 best moments regardless of where they fall in the runtime.
 
-TRANSCRIPT:
-{transcript_content[:15000]}
+FULL TRANSCRIPT:
+{transcript_content}
 
-Return a JSON array with exactly 10 segments following the format specified. Each segment must have accurate start and end times from the transcript."""
+Return ONLY the JSON array with exactly 10 segments ranked by engagement level."""
 
         print("Sending transcript to LLM for highlight extraction...")
         response = client.chat.completions.create(