update: prompt

app.py

@@ -441,18 +441,39 @@ def chat(
             link_start = "http"
         else:
             link_start = "https"
-        prompt = f"""You are a helpful assistant analyzing transcripts and generating timestamps and URL. Call ID is {cid}, Session ID is {rsid}, origin is {origin}, Call Type is {ct}.
-Transcript:\n{transcript_processor.get_transcript()}
-If a user asks timestamps for a specific topic, find the start time and end time of that specific topic and return answer in the format: 
-If the user provides a link to the agenda, use the correct_speaker_name_with_url function to correct the speaker names based on the agenda.
-If the user provides the correct call type, use the correct_call_type function to correct the call type. Call Type for street interviews is 'si'.
-Answer format:
-Topic: Heading [Timestamp: start_time - end_time]({link_start}://{{origin}}/collab/{{cid}}/{{rsid}}?st={{start_time_in_sec}}&et={{end_time_in_sec}}&uid={{uid}}"').
+        speaker_mapping = transcript_processor.speaker_mapping
+        prompt = f"""You are a helpful assistant analyzing transcripts and generating timestamps and URL. The user will ask you questions regarding the social media clips from the transcript.
+Call ID is {cid},
+Session ID is {rsid},
+origin is {origin},
+Call Type is {ct}.
+Speakers: {", ".join(speaker_mapping.values())}
+Transcript: {transcript_processor.get_transcript()}
 
+If a user asks timestamps for a specific topic or things, find the start time and end time of that specific topic and return answer in the format:
+Answers and URLs should be formated as follows:
+[Topic title <div id='topic' style="display: inline"> 22s at 12:30 </div>]({link_start}://{{origin}}/collab/{{cid}}/{{rsid}}?st={{750}}&et={{772}}&uid={{uid}})
 For Example:
 If the start time is 10:13 and end time is 10:18, the url will be:
 {link_start}://roll.ai/colab/1234aq_12314/51234151?st=613&et=618&uid=82314
 In the URL, make sure that after RSID there is ? and then rest of the fields are added via &.
+You can include multiple links here that can related to the user answer. ALWAYS ANSWER FROM THE TRANSCRIPT.
+
+Example 1:
+User: Suggest me some clips that can go viral on Instagram.
+Response:
+1. [Clip 1 <div id='topic' style="display: inline"> 22s at 12:30 </div>]({link_start}://{{origin}}/collab/{{cid}}/{{rsid}}?st={{750}}&et={{772}}&uid={{uid}})
+2. [Clip 2 <div id='topic' style="display: inline"> 10s at 10:00 </div>]({link_start}://{{origin}}/collab/{{cid}}/{{rsid}}?st={{600}}&et={{610}}&uid={{uid}})
+
+Example 2:
+User: Give me the URL where each person has introduced themselves.
+Response (Please provide exact time where the person has started introducing themselves. Normally person start introducing themselves by saying Hi, Hello, I am, My name is, etc.):
+1. [Person Name1 <div id='topic' style="display: inline"> 43s at 14:45 </div>]({link_start}://{{origin}}/collab/{{cid}}/{{rsid}}?st={{885}}&et={{928}}&uid={{uid}})
+2. [Person Name2 <div id='topic' style="display: inline"> 58s at 16:20 </div>]({link_start}://{{origin}}/collab/{{cid}}/{{rsid}}?st={{980}}&et={{1038}}&uid={{uid}})
+....
+
+If the user provides a link to the agenda, use the correct_speaker_name_with_url function to correct the speaker names based on the agenda.
+If the user provides the correct call type, use the correct_call_type function to correct the call type. Call Type for street interviews is 'si'.
 """
         messages = [{"role": "system", "content": prompt}]
 
@@ -466,7 +487,11 @@ In the URL, make sure that after RSID there is ? and then rest of the fields are
         messages.append({"role": "user", "content": message})
 
         completion = client.chat.completions.create(
-            model="gpt-4o-mini", messages=messages, tools=tools, stream=True
+            model="gpt-4o-mini",
+            messages=messages,
+            tools=tools,
+            stream=True,
+            temperature=0.5,
         )
         collected_messages = []
         tool_calls_detected = False
@@ -762,7 +787,7 @@ def create_chat_interface():
                 transcript_data = get_transcript_for_url(turl)
                 transcript_processor = TranscriptProcessor(
                     transcript_data=transcript_data,
-                    max_segment_duration=10,
+                    max_segment_duration=20 if ct != "si" else 10,
                     call_type=ct,
                 )