updates

.gitignore

@@ -1,2 +1,4 @@
 .env
 .venv
+/__pycache__
+/chroma_db

agent.py

@@ -1,7 +1,8 @@
+import json
 import os
+import re
 
 from dotenv import load_dotenv
-from langchain_community.vectorstores import Chroma
 from langchain_core.messages import HumanMessage, SystemMessage, ToolMessage
 from langchain_huggingface import (ChatHuggingFace, HuggingFaceEmbeddings,
                                    HuggingFaceEndpoint)
@@ -12,13 +13,11 @@ from tools import (absolute, add, analyze_csv_file, analyze_excel_file,
                    arvix_search, audio_transcription, compound_interest,
                    convert_temperature, divide, exponential,
                    extract_text_from_image, factorial, floor_divide,
-                   get_current_time_in_timezone,
-                   get_max_bird_species_count_from_video,
-                   greatest_common_divisor, is_prime, least_common_multiple,
-                   logarithm, modulus, multiply, percentage_calculator, power,
-                   python_code_parser, reverse_sentence,
-                   roman_calculator_converter, square_root, subtract,
-                   web_content_extract, web_search, wiki_search)
+                   get_current_time_in_timezone, greatest_common_divisor,
+                   is_prime, least_common_multiple, logarithm, modulus,
+                   multiply, percentage_calculator, power, python_code_parser,
+                   reverse_sentence, roman_calculator_converter, square_root,
+                   subtract, web_content_extract, web_search, wiki_search)
 
 # Load Constants
 load_dotenv()
@@ -34,8 +33,7 @@ tools = [
     is_prime, least_common_multiple, percentage_calculator,
     wiki_search, analyze_excel_file, arvix_search,
     audio_transcription, python_code_parser, analyze_csv_file,
-    extract_text_from_image, reverse_sentence, web_content_extract,
-    get_max_bird_species_count_from_video
+    extract_text_from_image, reverse_sentence, web_content_extract
 ]
 
 # Load system prompt
@@ -47,54 +45,13 @@ If you are asked for a number, don't use a comma to write your number, nor use u
 If you are asked for a string, don't use articles, neither abbreviations (e.g. for cities), and write the digits in plain text unless specified otherwise.
 If you are asked for a comma separated list, apply the above rules depending on whether the element to be put in the list is a number or a string.
 Format your output as: Answers (answers): [{"task_id": ..., "submitted_answer": ...}]
+Do not repeat the format or include any nested JSON. Output only one flat list as: Answers (answers): [{...}]
 """
 
-
 # System message
 sys_msg = SystemMessage(content=system_prompt)
 
 
-def get_vector_store(persist_directory="chroma_db"):
-    """
-    Initializes and returns a Chroma vector store.
-    If the database exists, it loads it. If not, it creates it,
-    adds some initial documents, and persists them.
-    """
-    embedding_function = HuggingFaceEmbeddings(model_name="sentence-transformers/all-mpnet-base-v2")
-
-    if os.path.exists(persist_directory) and os.listdir(persist_directory):
-        print("Loading existing vector store...")
-        vector_store = Chroma(
-            persist_directory=persist_directory,
-            embedding_function=embedding_function
-        )
-    else:
-        print("Creating new vector store...")
-        os.makedirs(persist_directory, exist_ok=True)
-        # Example documents to add
-        initial_documents = [
-            "The Principle of Double Effect is an ethical theory that distinguishes between the intended and foreseen consequences of an action.",
-            "St. Thomas Aquinas is often associated with the development of the Principle of Double Effect.",
-            "LangGraph is a library for building stateful, multi-actor applications with LLMs.",
-            "Chroma is a vector database used for storing and retrieving embeddings."
-        ]
-        vector_store = Chroma.from_texts(
-            texts=initial_documents,
-            embedding=embedding_function,
-            persist_directory=persist_directory
-        )
-        # No need to call persist() when using from_texts with a persist_directory
-
-    return vector_store
-
-# --- Initialize Vector Store and Retriever ---
-vector_store = get_vector_store()
-retriever_component = vector_store.as_retriever(
-    search_type="mmr",  # Use Maximum Marginal Relevance for diverse results
-    search_kwargs={'k': 2, 'lambda_mult': 0.5}  # Retrieve 2 documents
-)
-
-
 def build_graph():
     """Build the graph"""
     # First create the HuggingFaceEndpoint
@@ -127,32 +84,12 @@ def build_graph():
         formatted = f'Answers (answers): [{{"task_id": "{task_id}", "submitted_answer": "{answer_text}"}}]'
         return {"messages": [formatted]}
 
-    def retriever_node(state: MessagesState):
-        """
-        Retrieves relevant documents from the vector store based on the latest human message.
-        """
-        last_human_message = state["messages"][-1].content
-        retrieved_docs = retriever_component.invoke(last_human_message)
-        
-        if retrieved_docs:
-            retrieved_context = "\n\n".join([doc.page_content for doc in retrieved_docs])
-            # Create a ToolMessage to hold the retrieved context
-            context_message = ToolMessage(
-                content=f"Retrieved context from vector store:\n\n{retrieved_context}",
-                tool_call_id="retriever" # A descriptive ID
-            )
-            return {"messages": [context_message]}
-        
-        return {"messages": []}
-
     # --- Graph Definition ---
     builder = StateGraph(MessagesState)
-    # builder.add_node("retriever", retriever_node)
     builder.add_node("assistant", assistant)
     builder.add_node("tools", ToolNode(tools))
 
     builder.add_edge(START, "assistant")
-    # builder.add_edge("retriever", "assistant")
     builder.add_conditional_edges("assistant", tools_condition)
     builder.add_edge("tools", "assistant")
 
@@ -160,6 +97,30 @@ def build_graph():
     return builder.compile()
 
 
+def is_valid_agent_output(output):
+    """
+    Checks if the output matches the required format:
+    Answers (answers): [{"task_id": ..., "submitted_answer": ...}]
+    """
+    # Basic regex to check the format
+    pattern = r'^Answers \(answers\): \[(\{.*\})\]$'
+    match = re.match(pattern, output.strip())
+    if not match:
+        return False
+
+    # Try to parse the JSON part
+    try:
+        answers_list = json.loads(f'[{match.group(1)}]')
+        # Check required keys
+        for ans in answers_list:
+            if not isinstance(ans, dict):
+                return False
+            if "task_id" not in ans or "submitted_answer" not in ans:
+                return False
+        return True
+    except Exception:
+        return False
+
 # test
 if __name__ == "__main__":
     question = "When was a picture of St. Thomas Aquinas first added to the Wikipedia page on the Principle of double effect?"
@@ -168,8 +129,8 @@ if __name__ == "__main__":
     # Run the graph
     messages = [HumanMessage(content=question)]
     # The initial state for the graph
-    initial_state = {"messages": messages}
-    
+    initial_state = {"messages": messages, "task_id": "test123"}
+
     # Invoke the graph stream to see the steps
     for s in graph.stream(initial_state, stream_mode="values"):
         message = s["messages"][-1]
@@ -178,5 +139,9 @@ if __name__ == "__main__":
             print(message.content)
             print("-----------------------")
         else:
-            message.pretty_print()
-
+            output = str(message)
+            print("Agent Output:", output)
+            if is_valid_agent_output(output):
+                print("✅ Output is in the correct format!")
+            else:
+                print("❌ Output is NOT in the correct format!")

app.py

@@ -54,38 +54,8 @@ class BasicAgent:
         # The answer is expected to be in the 'content' of the last message.
         answer = response_messages['messages'][-1].content
         print(f"Agent full response: {answer}")
-
-        final_answer = ""
-        if not messages:
-            # print(f"No messages found in the result state for task {task_id}.")
-            return "AGENT ERROR: No messages returned by the agent."
-
-        for msg in reversed(messages):
-            if hasattr(msg, "content") and msg.content:
-                content = msg.content
-                if isinstance(content, str):
-                    if "FINAL ANSWER:" in content:
-                        final_answer = content.split("FINAL ANSWER:", 1)[1].strip()
-                        break 
-                    elif isinstance(msg, AIMessage):
-                        # If it's an AIMessage and no "FINAL ANSWER:" has been found yet,
-                        # tentatively set it. This will be overridden if a "FINAL ANSWER:" is found later.
-                        if not final_answer: 
-                            final_answer = content
-        
-        # If after checking all messages, final_answer is still from a non-"FINAL ANSWER:" AIMessage, that's our best guess.
-        # If final_answer is empty, it means no AIMessage with content or "FINAL ANSWER:" was found.
-        if not final_answer: # This means no "FINAL ANSWER:" and no AIMessage content was suitable
-            final_answer = "AGENT ERROR: Could not extract a final answer from the agent's messages."
-            # print(f"Could not extract final answer for task {task_id}. Messages: {messages}")
-
-        # print(f"FinalAgent returning answer for task_id '{task_id}': {final_answer[:100]}...")
-        print(f"FinalAgent returning answer: {final_answer[:100]}...")
-        return final_answer
-
-#        return answer
-        
-
+        return answer
+      
 
 def run_and_submit_all( profile: gr.OAuthProfile | None):
     """

old-tools.py

@@ -1,3 +1,9 @@
+import tempfile
+
+import cv2
+import torch
+from pytube import YouTube
+
 
 @tool
 def web_search(query: str) -> str:
@@ -23,3 +29,68 @@ def web_search(query: str) -> str:
         return results
     except Exception as e:
         return f"Error performing web search: {str(e)}"
+
+
+@tool
+def get_max_bird_species_count_from_video(url: str) -> Dict:
+    """
+    Downloads a YouTube video and returns the maximum number of unique bird species 
+    visible in any frame, along with the timestamp.
+
+    Parameters:
+        url (str): YouTube video URL
+
+    Returns:
+        dict: {
+            "max_species_count": int,
+            "timestamp": str,
+            "species_list": List[str],
+        }
+    """
+    # 1. Download YouTube video
+    yt = YouTube(url)
+    stream = yt.streams.filter(file_extension='mp4').get_highest_resolution()
+    temp_video_path = os.path.join(tempfile.gettempdir(), "video.mp4")
+    stream.download(filename=temp_video_path)
+
+    # 2. Load object detection model for bird species
+    # Load a fine-tuned YOLOv5 model or similar pretrained on bird species
+    model = torch.hub.load('ultralytics/yolov5', 'custom', path='best_birds.pt')  # path to your trained model
+
+    # 3. Process video frames
+    cap = cv2.VideoCapture(temp_video_path)
+    fps = cap.get(cv2.CAP_PROP_FPS)
+    frame_interval = int(fps * 1)  # 1 frame per second
+
+    max_species_count = 0
+    max_species_frame_time = 0
+    species_at_max = []
+
+    frame_idx = 0
+    while cap.isOpened():
+        ret, frame = cap.read()
+        if not ret:
+            break
+        if frame_idx % frame_interval == 0:
+            # Run detection
+            results = model(frame)
+            detected_species = set()
+            for *box, conf, cls in results.xyxy[0]:
+                species_name = model.names[int(cls)]
+                detected_species.add(species_name)
+
+            if len(detected_species) > max_species_count:
+                max_species_count = len(detected_species)
+                max_species_frame_time = int(cap.get(cv2.CAP_PROP_POS_MSEC)) // 1000
+                species_at_max = list(detected_species)
+
+        frame_idx += 1
+
+    cap.release()
+    os.remove(temp_video_path)
+
+    return {
+        "max_species_count": max_species_count,
+        "timestamp": f"{max_species_frame_time}s",
+        "species_list": species_at_max
+    }

requirements.txt

@@ -10,9 +10,7 @@ langchain-core
 langchain-community
 langgraph
 langchain-huggingface
-langchain-chroma
-chromadb # Explicitly add the Chroma database
-sentence-transformers
+# sentence-transformers
 langfuse
 langchain-google-genai
 langchain-tavily
@@ -40,18 +38,18 @@ typing-extensions
 #tenacity
 # loguru
 
-torch
-torchvision
-opencv-python
-pytube
+# torch
+# torchvision
+# opencv-python
+# pytube
 
 # YOLOv5 and dependencies
-numpy
-matplotlib
-scipy
-seaborn
-tqdm
-pyyaml
-pillow
+# numpy
+# matplotlib
+# scipy
+# seaborn
+# tqdm
+# pyyaml
+# pillow
 
 # git+https://github.com/ultralytics/yolov5.git

tools.py

@@ -2,13 +2,10 @@ import base64
 import datetime
 import math
 import os
-import tempfile
 from typing import Dict, Union
 
-import cv2
 import pandas
 import pytz
-import torch
 from bs4 import BeautifulSoup
 from langchain_community.document_loaders import (
     ArxivLoader, AssemblyAIAudioTranscriptLoader, WikipediaLoader)
@@ -19,7 +16,6 @@ from langchain_core.messages import HumanMessage
 from langchain_core.tools import tool
 from langchain_google_genai import ChatGoogleGenerativeAI
 from langchain_tavily import TavilySearch
-from pytube import YouTube
 
 
 @tool
@@ -742,70 +738,6 @@ def reverse_sentence(text: str) -> str:
     """
     return text[::-1]
 
-@tool
-def get_max_bird_species_count_from_video(url: str) -> Dict:
-    """
-    Downloads a YouTube video and returns the maximum number of unique bird species 
-    visible in any frame, along with the timestamp.
-
-    Parameters:
-        url (str): YouTube video URL
-
-    Returns:
-        dict: {
-            "max_species_count": int,
-            "timestamp": str,
-            "species_list": List[str],
-        }
-    """
-    # 1. Download YouTube video
-    yt = YouTube(url)
-    stream = yt.streams.filter(file_extension='mp4').get_highest_resolution()
-    temp_video_path = os.path.join(tempfile.gettempdir(), "video.mp4")
-    stream.download(filename=temp_video_path)
-
-    # 2. Load object detection model for bird species
-    # Load a fine-tuned YOLOv5 model or similar pretrained on bird species
-    model = torch.hub.load('ultralytics/yolov5', 'custom', path='best_birds.pt')  # path to your trained model
-
-    # 3. Process video frames
-    cap = cv2.VideoCapture(temp_video_path)
-    fps = cap.get(cv2.CAP_PROP_FPS)
-    frame_interval = int(fps * 1)  # 1 frame per second
-
-    max_species_count = 0
-    max_species_frame_time = 0
-    species_at_max = []
-
-    frame_idx = 0
-    while cap.isOpened():
-        ret, frame = cap.read()
-        if not ret:
-            break
-        if frame_idx % frame_interval == 0:
-            # Run detection
-            results = model(frame)
-            detected_species = set()
-            for *box, conf, cls in results.xyxy[0]:
-                species_name = model.names[int(cls)]
-                detected_species.add(species_name)
-
-            if len(detected_species) > max_species_count:
-                max_species_count = len(detected_species)
-                max_species_frame_time = int(cap.get(cv2.CAP_PROP_POS_MSEC)) // 1000
-                species_at_max = list(detected_species)
-
-        frame_idx += 1
-
-    cap.release()
-    os.remove(temp_video_path)
-
-    return {
-        "max_species_count": max_species_count,
-        "timestamp": f"{max_species_frame_time}s",
-        "species_list": species_at_max
-    }
-
 @tool
 def web_search(query: str) -> str:
     """