LE prompt with reasoning

app.py

@@ -1,7 +1,7 @@
 import json
 import os
 from dataclasses import dataclass
-from typing import Dict, List
+from typing import Dict, Generator, List
 
 import gradio as gr
 import requests
@@ -388,7 +388,7 @@ def get_transcript_for_url(url: str) -> dict:
 
 def get_initial_analysis(
     transcript_processor: TranscriptProcessor, cid, rsid, origin, ct, uid
-) -> str:
+) -> Generator[str, None, None]:
     """Perform initial analysis of the transcript using OpenAI."""
     try:
         transcript = transcript_processor.get_transcript()
@@ -489,8 +489,65 @@ CRITICAL: When analyzing timestamps, you must verify that in the duration specif
 2. The speaker talks continuously for at least 20 seconds
 3. The clip ends BEFORE any interruption or speaker change
 """
+            print(" , ".join(speaker_mapping.values()))
+            reasoning_prompt = f"""For each Speaker {" , ".join(speaker_mapping.values())}
+in the transcript: {transcript}
+
+Your job is to generate the thinking about the short social media clips for each speaker where they discuss. Think step by step and return a JSON at the end of the thinking.
+Generate the thinking for atleast 2 clips for each speaker.
+Return Format:
+- Name of the Speaker
+- Detailed Step by Step Thinking for each speaker from thier content and the topic they are talking about
+After you have completed the thinking, give me a JSON of the thinking.
+```json
+[
+    {{
+        "Speaker 0": [
+            {{
+                "Topic Title": "...",
+                "Starting Sentence of that speaker": "...",
+                "Ending Sentence where the topic ends": "...."
+            }},
+            {{
+                "Topic Title": "...",
+                "Starting Sentence of that speaker": "....",
+                "Ending Sentence of that speaker where the topic ends": "....."
+            }}
+        ]
+    }},
+    {{
+        "Speaker 1": [
+            {{
+                "Topic Title": "....",
+                "Starting Sentence of that speaker": ".....",
+                "Ending Sentence of that speaker": "....."
+            }},
+            {{
+                "Topic Title": "......",
+                "Starting Sentence of that speaker": "....",
+                "Ending Sentence of that speaker": "....."
+            }}
+        ]
+    }},
+    ....
+]
+```
+"""
+            thinking_completion = client.chat.completions.create(
+                model="gpt-4o",
+                messages=[
+                    {"role": "system", "content": reasoning_prompt},
+                ],
+                stream=False,
+                temperature=0.4,
+            )
+            thinking = thinking_completion.choices[0].message.content
+            print("Thinking is:\n", thinking)
+            thinking_json = thinking[thinking.find("{") : thinking.rfind("}") + 1]
+
+            user_prompt = f"""User ID: {uid}
+Intelligent Thinking Context: {thinking_json}
 
-            user_prompt = f"""User ID: {uid}\n
 Your task is to generate the social media clips following these strict rules:
 
 1. TIMESTAMP SELECTION:
@@ -517,7 +574,6 @@ SPEAKER FORMAT:
 **Speaker Name**
 ....
 """
-            print(user_prompt, speaker_mapping)
 
             completion = client.chat.completions.create(
                 model="gpt-4o",