Update app.py

app.py

@@ -16,41 +16,59 @@ openai_key = os.environ.get("OPENAI_API_KEY")
 search_tool = DuckDuckGoSearchTool()
 
 ##Tool 1
-import re
-from youtube_transcript_api import YouTubeTranscriptApi
-from smolagents import Tool
-
 from smolagents import Tool
 from youtube_transcript_api import YouTubeTranscriptApi
 from urllib.parse import urlparse, parse_qs
 
-class YouTubeTranscriptTool(Tool):
-    name = "youtube_transcript"
-    description = "Fetches the full transcript of a YouTube video from its URL."
+class YouTubeDialogueTool(Tool):
+    name = "youtube_dialogue_qa"
+    description = "Extracts the transcript and finds what Teal'c says in response to a specific question."
 
     inputs = {
         "url": {
             "type": "string",
-            "description": "The full YouTube video URL"
+            "description": "Full YouTube video URL"
+        },
+        "question": {
+            "type": "string",
+            "description": 'Exact question asked (e.g., "Isn\'t that hot?")'
+        },
+        "speaker": {
+            "type": "string",
+            "description": "Name of the character who responds (e.g., 'Teal'c')"
         }
     }
 
     output_type = "string"
 
-    def forward(self, url: str) -> str:
+    def forward(self, url: str, question: str, speaker: str = "Teal'c") -> str:
         try:
-            # Extract video ID from URL
+            # Extract video ID
             parsed = urlparse(url)
             video_id = parse_qs(parsed.query).get("v", [None])[0]
-
             if not video_id:
-                return "Error: Invalid YouTube URL or missing video ID."
+                return "ERROR: Invalid YouTube URL or missing video ID."
+
+            # Get transcript
+            transcript = YouTubeTranscriptApi.get_transcript(video_id)
+
+            # Build list of lines
+            lines = []
+            for entry in transcript:
+                text = entry["text"].replace("\n", " ").strip()
+                if text:
+                    lines.append(text)
+
+            # Find the question and capture the next line(s)
+            for idx, line in enumerate(lines):
+                if question.lower().rstrip("?") in line.lower():
+                    # Look ahead for a line that matches speaker response style
+                    if idx + 1 < len(lines):
+                        return lines[idx + 1].strip()
+                    break
 
-            # Fetch the transcript
-            transcript_list = YouTubeTranscriptApi.get_transcript(video_id)
-            transcript_text = " ".join(entry["text"] for entry in transcript_list)
+            return "No response found in transcript."
 
-            return transcript_text[:5000]  # Optional: truncate to 5000 chars
         except Exception as e:
             return f"Error retrieving transcript: {str(e)}"
 
@@ -115,7 +133,7 @@ class WikipediaQATool(Tool):
 
 wiki_tool = WikipediaQATool()
 #excel_tool = ExcelAnalysisTool()
-yt_tool = YouTubeTranscriptTool()
+yt_tool = YouTubeDialogueTool()
 
 async def run_and_submit_all(profile: gr.OAuthProfile | None):
     log_output = ""
@@ -127,7 +145,7 @@ async def run_and_submit_all(profile: gr.OAuthProfile | None):
                 model=OpenAIServerModel(model_id="gpt-4o-mini", 
                                         api_key=os.environ["OPENAI_API_KEY"], 
                                        temperature=0.0),
-            max_steps=4,
+            max_steps=7,
             verbosity_level=2
         )
     except Exception as e: