tools.py

import pandas as pd 
from smolagents import tool 
import litellm 
from dotenv import load_dotenv 
import os 
import whisper
from pathlib import Path

@tool 
def image_interpretation(file:str) ->str:
    """ 
    Returns a high quality textual description of a provided image 

    Args:
        file: A .png or .jpg image provided by the user 
    Returns:
        Text describing the scene 
    Example:
        query 1: dog.png 
        output 1: A golden retriever sitting on the beach at mid-day

        query 2: scene.jpg
        output 2: A landscape picture of sunrise at a mountain range
    """
    load_dotenv()
    key = os.getenv("GROQ_API_KEY")
    prompt = f""" 
        Open the image at {file} and analyze it and understand it. 
        Generate a high quality textual description of what you see, it has to be detailed
        and accurate, don't make up anything you don't see in the image. 
        Return ONLY the textual description. 
    """
    message = [{'role':'user', 'content':prompt}]
    response = litellm.completion(model="groq/llama-3.3-70b-versatile", messages=message, api_key=key)
    description = response.choices[0].message.content
    return description

@tool
def create_transcripts(file:str) -> str:
    """
    Returns high quality transcripts of audio files

    Args:
        file: Users .mp3 file containing audio
    Returns:
        Text file (transcript of conversation) 
    
    Example: 
        file: hello_world.mp3
        output: "Hello world, I am a computer"
    """

    file_path = Path(file).resolve()
    model = whisper.load_model("base")
    result = model.transcribe(str(file_path))
    return result["text"]

@tool 
def file_handling(file:str) -> pd.DataFrame:
    """
    Returns excel sheets and csv files converted into a pandas Dataframe  

    Args:
        file: The file path for the 
    Outputs: 
        The user's original content but as a pandas dataframe 
    """
    extension = file.split('.')[-1]
    if extension == 'xlsx':
        df = pd.read_excel(f"{file}")
    elif extension == 'csv':
        df = pd.read_csv(f"{file}")
    else:
        df = "Sorry, can't open any other files"
    return df 

@tool
def fix_input(query:str) -> str:
    """
    Handles incomplete, missing or jumbled user input

    Args:
        query: User query that can potentially contain formatting errors like missing errors, 
        spelling mistakes, jumbled word letter, etc. 
    Returns:
        The corrected user input, which can be interpreted properly in natural language 
    
    Example: 
        query 1: What is th capirsl o Franc
        output 1: What is the capital of France? 

        query 2: I am planning a road trip with my family of 4. Which should I rent? 
        output 2: I am planning a road trip with my family of 4. Which car should I rent? 
    """
    load_dotenv()
    key = os.getenv("GROQ_API_KEY")
    prompt = f""" Is the user's question: {query} in the correct natural language format, is it clear? 
    Can you understand it as it is? If yes, then return the same query. If not, try (i) correcting 
    spelling mistakes (ii) filling in missing letters / words (iii) re-arranging jumbled words / word orders 
    (iv) reversing the string (v) any other suitable string and sentence operations. 
    Finally return ONLY the corrected string.   
    """
    message = [{"role": "user", "content": prompt}]
    response = litellm.completion(model="groq/llama-3.3-70b-versatile", messages=message, api_key=key)
    corrected_query = response.choices[0].message.content
    return corrected_query