agent.py

# agent.py

import os
from dotenv import load_dotenv
from langgraph.graph import START, StateGraph, MessagesState
from langgraph.prebuilt import tools_condition
from langgraph.prebuilt import ToolNode
from langchain_google_genai import ChatGoogleGenerativeAI
from langchain_groq import ChatGroq
from langchain_huggingface import ChatHuggingFace, HuggingFaceEndpoint, HuggingFaceEmbeddings
from langchain_community.tools.tavily_search import TavilySearchResults
from langchain_community.document_loaders import WikipediaLoader
from langchain_community.document_loaders import ArxivLoader
from langchain_community.vectorstores import SupabaseVectorStore
from langchain_core.messages import SystemMessage, HumanMessage
from langchain_core.tools import tool
from langchain.tools.retriever import create_retriever_tool
from supabase.client import Client, create_client
from sentence_transformers import SentenceTransformer
from langchain.embeddings.base import Embeddings
from typing import List
import numpy as np
import yaml

import pandas as pd
import uuid
import requests
import json
from langchain_core.documents import Document
from langchain_community.vectorstores import FAISS
from langchain_community.embeddings import HuggingFaceEmbeddings

from youtube_transcript_api import YouTubeTranscriptApi
from youtube_transcript_api._errors import TranscriptsDisabled, VideoUnavailable
import re

from langchain_community.document_loaders import TextLoader, PyMuPDFLoader
from docx import Document as DocxDocument
import openpyxl
from io import StringIO

from transformers import BertTokenizer, BertModel
import torch



load_dotenv()

@tool
def multiply(a: int, b: int) -> int:
    """Multiply two numbers.
    Args:
        a: first int
        b: second int
    """
    return a * b

@tool
def add(a: int, b: int) -> int:
    """Add two numbers.
    
    Args:
        a: first int
        b: second int
    """
    return a + b

@tool
def subtract(a: int, b: int) -> int:
    """Subtract two numbers.
    
    Args:
        a: first int
        b: second int
    """
    return a - b

@tool
def divide(a: int, b: int) -> int:
    """Divide two numbers.
    
    Args:
        a: first int
        b: second int
    """
    if b == 0:
        raise ValueError("Cannot divide by zero.")
    return a / b

@tool
def modulus(a: int, b: int) -> int:
    """Get the modulus of two numbers.
    
    Args:
        a: first int
        b: second int
    """
    return a % b

@tool
def wiki_search(query: str) -> str:
    """Search Wikipedia for a query and return maximum 2 results.
    
    Args:
        query: The search query."""
    search_docs = WikipediaLoader(query=query, load_max_docs=2).load()
    formatted_search_docs = "\n\n---\n\n".join(
        [
            f'<Document source="{doc.metadata["source"]}" page="{doc.metadata.get("page", "")}"/>\n{doc.page_content}\n</Document>'
            for doc in search_docs
        ])
    return {"wiki_results": formatted_search_docs}

@tool
def web_search(query: str) -> str:
    """Search Tavily for a query and return maximum 3 results.
    
    Args:
        query: The search query."""
    search_docs = TavilySearchResults(max_results=3).invoke(query=query)
    formatted_search_docs = "\n\n---\n\n".join(
        [
            f'<Document source="{doc.metadata["source"]}" page="{doc.metadata.get("page", "")}"/>\n{doc.page_content}\n</Document>'
            for doc in search_docs
        ])
    return {"web_results": formatted_search_docs}

@tool
def arvix_search(query: str) -> str:
    """Search Arxiv for a query and return maximum 3 result.
    
    Args:
        query: The search query."""
    search_docs = ArxivLoader(query=query, load_max_docs=3).load()
    formatted_search_docs = "\n\n---\n\n".join(
        [
            f'<Document source="{doc.metadata["source"]}" page="{doc.metadata.get("page", "")}"/>\n{doc.page_content[:1000]}\n</Document>'
            for doc in search_docs
        ])
    return {"arvix_results": formatted_search_docs}



@tool
def analyze_attachment(file_path: str) -> str:
    """
    Analyzes attachments including PDF, TXT, DOCX, and XLSX files and returns text content.

    Args:
        file_path: Local path to the attachment.
    """
    if not os.path.exists(file_path):
        return f"File not found: {file_path}"

    if file_path.lower().endswith(".pdf"):
        loader = PyMuPDFLoader(file_path)
        documents = loader.load()
        content = "\n\n".join([doc.page_content for doc in documents])

    elif file_path.lower().endswith(".txt"):
        loader = TextLoader(file_path)
        documents = loader.load()
        content = "\n\n".join([doc.page_content for doc in documents])

    elif file_path.lower().endswith(".docx"):
        doc = DocxDocument(file_path)
        content = "\n".join([para.text for para in doc.paragraphs])

    elif file_path.lower().endswith(".xlsx"):
        wb = openpyxl.load_workbook(file_path, data_only=True)
        content = ""
        for sheet in wb:
            content += f"Sheet: {sheet.title}\n"
            for row in sheet.iter_rows(values_only=True):
                content += "\t".join([str(cell) if cell is not None else "" for cell in row]) + "\n"

    else:
        return "Unsupported file format. Please use PDF, TXT, DOCX, or XLSX."

    return content[:3000]  # Limit size for readability


@tool
def get_youtube_transcript(url: str) -> str:
    """
    Fetch transcript text from a YouTube video.
    
    Args:
        url (str): Full YouTube video URL.
    
    Returns:
        str: Transcript text as a single string.
    
    Raises:
        ValueError: If no transcript is available or URL is invalid.
    """
    try:
        # Extract video ID
        video_id = extract_video_id(url)
        transcript = YouTubeTranscriptApi.get_transcript(video_id)

        # Combine all transcript text
        full_text = " ".join([entry['text'] for entry in transcript])
        return full_text

    except (TranscriptsDisabled, VideoUnavailable) as e:
        raise ValueError(f"Transcript not available: {e}")
    except Exception as e:
        raise ValueError(f"Failed to fetch transcript: {e}")

@tool
def extract_video_id(url: str) -> str:
    """
    Extract the video ID from a YouTube URL.
    """
    match = re.search(r"(?:v=|youtu\.be/)([A-Za-z0-9_-]{11})", url)
    if not match:
        raise ValueError("Invalid YouTube URL")
    return match.group(1)




# -----------------------------
# Load configuration from YAML
# -----------------------------
with open("config.yaml", "r") as f:
    config = yaml.safe_load(f)

provider = config["provider"]
model_config = config["models"][provider]

#prompt_path = config["system_prompt_path"]
enabled_tool_names = config["tools"]


# -----------------------------
# Load system prompt
# -----------------------------
# load the system prompt from the file
with open("system_prompt.txt", "r", encoding="utf-8") as f:
    system_prompt = f.read()

# System message
sys_msg = SystemMessage(content=system_prompt)


# -----------------------------
# Map tool names to functions
# -----------------------------
tool_map = {
    "multiply": multiply,
    "add": add,
    "subtract": subtract,
    "divide": divide,
    "modulus": modulus,
    "wiki_search": wiki_search,
    "web_search": web_search,
    "arvix_search": arvix_search,
    "get_youtube_transcript": get_youtube_transcript,
    "extract_video_id": extract_video_id,
    "analyze_attachment": analyze_attachment,
    
}

tools = [tool_map[name] for name in enabled_tool_names]


# -------------------------------
# Step 2: Load the JSON file or tasks (Replace this part if you're loading tasks dynamically)
# -------------------------------
# Here we assume the tasks are already fetched from a URL or file.
# For now, using an example JSON array directly. Replace this with the actual loading logic.

tasks = [
    {
        "task_id": "8e867cd7-cff9-4e6c-867a-ff5ddc2550be",
        "question": "How many studio albums were published by Mercedes Sosa between 2000 and 2009 (included)? You can use the latest 2022 version of English Wikipedia.",
        "Level": "1",
        "file_name": ""
    },
    {
        "task_id": "a1e91b78-d3d8-4675-bb8d-62741b4b68a6",
        "question": "In the video https://www.youtube.com/watch?v=L1vXCYZAYYM, what is the highest number of bird species to be on camera simultaneously?",
        "Level": "1",
        "file_name": ""
    }
]

# -------------------------------
# Step 3: Create Documents from Each JSON Object
# -------------------------------
docs = []
for task in tasks:
    # Debugging: Print the keys of each task to ensure 'question' exists
    print(f"Keys in task: {task.keys()}")

    # Ensure the required field 'question' exists
    if 'question' not in task:
        print(f"Skipping task with missing 'question' field: {task}")
        continue

    content = task.get('question', "").strip()
    if not content:
        print(f"Skipping task with empty 'question': {task}")
        continue

    # Add unique ID to each document
    task['id'] = str(uuid.uuid4())

    # Create a document from the task data
    docs.append(Document(page_content=content, metadata=task))



# -------------------------------
# Step 4: Set up HuggingFace Embeddings and FAISS VectorStore
# -------------------------------
# Initialize HuggingFace Embedding model
#embedding_model = HuggingFaceEmbeddings(model_name="sentence-transformers/all-mpnet-base-v2")
#embedding_model = HuggingFaceEmbeddings(model_name="BAAI/bge-base-en-v1.5")



class BERTEmbeddings(Embeddings):
    def __init__(self, model_name='bert-base-uncased'):
        self.tokenizer = BertTokenizer.from_pretrained(model_name)
        self.model = BertModel.from_pretrained(model_name)
        self.model.eval()  # Set to evaluation mode

    def embed_documents(self, texts: List[str]) -> List[List[float]]:
        return self._embed(texts)

    def embed_query(self, text: str) -> List[float]:
        return self._embed([text])[0]

    def _embed(self, texts: List[str]) -> List[List[float]]:
        inputs = self.tokenizer(texts, return_tensors='pt', padding=True, truncation=True)
        with torch.no_grad():
            outputs = self.model(**inputs)
        embeddings = outputs.last_hidden_state.mean(dim=1)  # Mean pooling
        return embeddings.cpu().numpy().tolist()

# Example usage of BERTEmbedding with LangChain

embedding_model = BERTEmbeddings(model_name="bert-base-uncased")

# Sample text (replace with your own text)
docs = [
    Document(page_content="Mercedes Sosa released many albums between 2000 and 2009."),
    Document(page_content="She was a prominent Argentine folk singer."),
    Document(page_content="Her album 'Al Despertar' was released in 1998."),
    Document(page_content="She continued releasing music well into the 2000s.")
]
# Get the embeddings for the documents
vector_store = FAISS.from_documents(docs, embedding_model)

# Now, you can use the embeddings with FAISS or other retrieval systems
# For example, with FAISS:
from langchain.vectorstores import FAISS

# Assuming 'docs' contains your list of documents and 'embedding_model' is the model you created
vector_store = FAISS.from_documents(docs, embedding_model)
vector_store.save_local("faiss_index")


# -----------------------------
# Step 4: Create Retriever Tool
# -----------------------------
retriever = vector_store.as_retriever()

question_retriever_tool = create_retriever_tool(
    retriever=retriever,
    name="Question_Search",
    description="A tool to retrieve documents related to a user's question."
)



def retriever(state: MessagesState):
    """Retriever node using similarity scores for filtering"""
    query = state["messages"][0].content
    results = vector_store.similarity_search_with_score(query, k=4)  # top 4 matches

    # Filter by score (lower is more similar; adjust threshold as needed)
    threshold = 0.8
    filtered = [doc for doc, score in results if score < threshold]

    if not filtered:
        example_msg = HumanMessage(content="No relevant documents found.")
    else:
        content = "\n\n".join(doc.page_content for doc in filtered)
        example_msg = HumanMessage(
            content=f"Here are relevant reference documents:\n\n{content}"
        )

    return {"messages": [sys_msg] + state["messages"] + [example_msg]}



tools = [
    multiply,
    add,
    subtract,
    divide,
    modulus,
    wiki_search,
    web_search,
    arvix_search,
]


def get_llm(provider: str, config: dict):
    if provider == "google":
        return ChatGoogleGenerativeAI(model=config["model"], temperature=config["temperature"])
    elif provider == "groq":
        return ChatGroq(model=config["model"], temperature=config["temperature"])
    elif provider == "huggingface":
        return ChatHuggingFace(
            llm=HuggingFaceEndpoint(url=config["url"], temperature=config["temperature"])
        )
    else:
        raise ValueError(f"Invalid provider: {provider}")


# Build graph function
def build_graph():
    """Build the graph based on provider"""
    llm = get_llm(provider, model_config)
    llm_with_tools = llm.bind_tools(tools)

    # Node
    def assistant(state: MessagesState):
        """Assistant node"""
        return {"messages": [llm_with_tools.invoke(state["messages"])]}
    
    def retriever(state: MessagesState):
        user_query = state["messages"][0].content
        similar_docs = vector_store.similarity_search(user_query)
    
        if not similar_docs:
            print("No similar docs found in FAISS. Using wiki_search.")
            wiki_result = wiki_search.invoke(user_query)
            return {
                "messages": [
                    sys_msg,
                    state["messages"][0],
                    HumanMessage(content=f"Using Wikipedia search:\n\n{wiki_result['wiki_results']}")
                ]
            }
        else:
            return {
                "messages": [
                    sys_msg,
                    state["messages"][0],
                    HumanMessage(content=f"Reference question:\n\n{similar_docs[0].page_content}")
                ]
            }


    
    builder = StateGraph(MessagesState)
    builder.add_node("retriever", retriever)
    builder.add_node("assistant", assistant)
    builder.add_node("tools", ToolNode(tools))
    builder.add_edge(START, "retriever")
    builder.add_edge("retriever", "assistant")
    builder.add_conditional_edges(
        "assistant",
        tools_condition,
    )
    builder.add_edge("tools", "assistant")

    # Compile graph
    return builder.compile()