main.py

from fasthtml_hf import setup_hf_backup
import os
import pandas as pd
import traceback
from datetime import datetime
from typing import Literal
from pydantic import BaseModel, Field
from fasthtml.common import * 
from langchain_core.prompts import PromptTemplate
from langchain.output_parsers import PydanticOutputParser
from langchain_openai import ChatOpenAI
from langchain_anthropic import ChatAnthropic
from langchain_google_genai import ChatGoogleGenerativeAI
from langchain_community.utilities.wikipedia import WikipediaAPIWrapper
from langchain_community.tools.wikipedia.tool import WikipediaQueryRun

# Set up the app, including daisyui and tailwind for the chat component
tlink = Script(src="https://cdn.tailwindcss.com"),
dlink = Link(rel="stylesheet", href="https://cdn.jsdelivr.net/npm/[email protected]/dist/full.min.css")

assets_dir = "/Users/manaranjanp/Documents/Work/MyLearnings/fastHTML/llmtimeline/assets"

app = FastHTML(hdrs=(tlink, dlink, picolink))

# Pydantic models
class Event(BaseModel):
    time: datetime = Field(description="When the event occurred")
    description: str = Field(description="A summary of what happened. Not more than 20 words.")
    sentiment: Literal["Positive", "Negative"] = Field(..., description="Categorization of the event sentiment")

class EventResponse(BaseModel):
    events: List[Event] = Field(min_length=5, max_length=30, description="List of events extracted from the context")

# Set up the Pydantic output parser
parser = PydanticOutputParser(pydantic_object=EventResponse)

# LangChain prompt template with format instructions
event_extraction_template = """
Extract the time based informations or events from the context and return a list of events with time,  event description and event sentiment type whether it was positive or negative event.
The context may contain information about people, organization or any other entity. 

<context>
{context}
</context>

The response must follow the following schema strictly. There will be penalty for not following the schema.

<schema>
{format_instructions}
</schema>

Must ensure the event belongs to the topic {topic} and try to get at least {numevents} unique events possible from the context.

Output:
"""

event_prompt = PromptTemplate(
    input_variables=["topic", "context"],
    partial_variables={"format_instructions": parser.get_format_instructions()},
    template=event_extraction_template
)    

# Function to get the appropriate language model based on user selection
def getModel(model, key):
    if(model == 'OpenAI Gpt-4o'):
        os.environ['OPENAI_API_KEY'] = key
        return ChatOpenAI(temperature=0,  # Set to 0 for deterministic output
                    model="gpt-4o-2024-08-06",  # Using the GPT-4 Turbo model
                    max_tokens=8000)  # Limit the response length
    elif (model == 'Anthropic Claude'):
        os.environ['ANTHROPIC_API_KEY'] = key
        return ChatAnthropic(model='claude-3-5-sonnet-20240620')  # Limit the response length
    else:
        os.environ['GOOGLE_API_KEY'] = key
        return ChatGoogleGenerativeAI(
            model="gemini-1.5-pro",
            temperature=0,
            max_tokens=8000,
            max_retries=2,
        )

    

# Function to generate an HTML table from the summary object
#def generate_timeline_html(timeline):
#    rows = []
#    for idx, tline in timeline.iterrows(): 
#        if(tline['Sentiment'] == "Positive"):           
#            rows.append(Div(Div( H2(tline['Time']), P(tline['Event']), cls = 'content'), cls = "container left"))                
#        else:
#            rows.append(Div(Div( H2(tline['Time']), P(tline['Event']), cls = 'content'), cls = "container right"))                
#
#    return  Div(*rows, cls="timeline")


# Function to generate an HTML table from the summary object
def generate_timeline_html(timeline):
    rows = []
    
    for idx, tline in timeline.iterrows(): 
        if idx % 2 == 0:
            rows.append(Li(Div(File("./assets/circle.svg"), cls = "timeline-middle"), 
                        Div(Time(tline['TimeStr'], 
                                 cls = "font-mono italic"), 
                            Div(tline['Event'],
                                 cls = 'text-lg font-black'), 
                                 cls = "timeline-start mb-10 md:text-end"),                            
                        Hr()))        
        else:
            rows.append(Li(Div(File("./assets/circle.svg"), cls = "timeline-middle"), 
                        Div(Time(tline['TimeStr'], 
                                 cls = "font-mono italic"),
                             Div(tline['Event'], 
                                 cls = 'text-lg font-black'), 
                                 cls = "timeline-end mb-10"),                            
                        Hr()))                                                             
               
    return  Ul(*rows, cls="timeline timeline-vertical")


def get_timeline_df(result):
    
    results_data = []
    # Parse the final result into GradedQAPair objects
    try:        
        if not isinstance(result, EventResponse):
            raise ValueError(f"Expected a list, but got {type(result)}")
        
    except Exception as e:
        print(f"An error occurred during analysis: {str(e)}")
        raise
    
    except Exception as e:
        print(f"An error occurred during analysis: {str(e)}")
        raise     
            
    if isinstance(result, EventResponse):
        # Create a list to hold the data for the DataFrame
        
        for event in result.events:
            results_data.append({
                'Time': event.time,
                'Event': event.description,
                'Sentiment': event.sentiment
            })

    df = pd.DataFrame(results_data)      
    df = df.sort_values("Time", ascending = True).reset_index()
    df['TimeStr'] = df['Time'].map(lambda x: x.strftime('%d/%m/%Y'))

    return df  
        
# Placeholder function for Q&A generation
def generate_timeline(topic, numevents, llm):
    # This function will be implemented later
    # For now, return a sample DataFrame

#    titles = wikipedia.search(topic, results = 1)
#    page = wikipedia.page(titles[0])
#    wiki_content = page.content

    
    wikipedia = WikipediaQueryRun(api_wrapper=WikipediaAPIWrapper(top_k_results=3, doc_content_chars_max=5000))
    wiki_content = wikipedia.run(topic)

    print(f"wiki_content: {wiki_content}")
#    print(f"wiki_artifact: {wiki_artifact}")
    
    chain = event_prompt | llm | parser

    result = chain.invoke({"context" : wiki_content, 
                           "topic": topic, 
                           "numevents": numevents})
    
    try:
        # Parse the output using PydanticOutputParser
        # response = parser.parse(result)
        # Create the DataFrame

        print(f"Results: {result}")

 #       timeline = parser.parse(result)
        
        df = get_timeline_df(result)
                
        # Optionally, save the DataFrame to a CSV file
        df.to_csv(f"{topic.replace(' ', '_')}_timeline.csv", index=True)
        print("Results saved to 'results.csv'")    
    except Exception as e:
        print(f"Error parsing LLM output: {str(e)}")
        return None

    return df

# Function to generate the configuration form for the web interface
def getConfigForm():
    return Card(Form(hx_post="/submit", hx_target="#result", hx_swap_oob="innerHTML", hx_indicator="#indicator")(
            Div(
                Label(Strong("Model and Topic: "), style="color:#3498db; font-size:25px;")
            ),
            Div(
                Span(Strong('Model: '), cls ="badge"),
                Select(Option("OpenAI Gpt-4o"), Option("Anthropic Claude"), Option("Google Gemini"), id="model", cls = 'select w-full max-w-xs')
            ),
            Div(
                Span(Strong('API Key: '), cls ="badge"),
                Input(id="secret", type="password", placeholder="Key: "),
            ),
            Div(
                Span(Strong('Topic for timeline (Person/Organization/Event): '), cls ="badge"),
                Input(type = 'text', 
                      id="topic",
                      cls = "input w-full max-w-xs",
                      placeholder = "Type here")
            ),
            Div(
                Span(Strong('How many events: '), cls ="badge"),
                Select(Option("5"), Option("10"), Option("20"), Option("30"), id="numevents", cls = 'select w-full max-w-xs')
            ),
            Div(                
                Button("Generate Timeline", cls = 'btn')
            ),
            Div(              
                Br(),                  
                A("Developed by Manaranjan Pradhan", href="http://www.manaranjanp.com/", 
                  target="_blank", 
                  style = 'color: red; font-size: 16px;')                      
            )))

# Define the route for the homepage
@app.get('/')
def homepage():
    return Titled(Card(H2('Generate a Timeline Dashboard using AI', cls = 'text-4xl font-bold')),  Grid( getConfigForm(),
        Div(
            Div(id="result"),
            Div(Label(Strong('Generating timeline for the topic.... take a deep breath....')),
            Progress(), id="indicator", cls="htmx-indicator")
        )
        , style="grid-template-columns: 400px 1000px; gap: 50px;"
    ))

@app.get('/assets/{fname:path}.{ext}')
async def get(fname: str, ext: str):
    fpath:str = (assets_dir)+'/'+str(fname)+'.'+str(ext)
    if os.path.isfile(fpath):
        response = FileResponse(fpath, media_type="image/svg")
        print("file sent:"+fpath)
    else:
        print("file failed:"+fpath)
        response = HTTPException(status_code=404, detail="File not found")

# Define the route for form submission
@app.post('/submit')
async def post(d:dict):
    try:            
        # Get the appropriate language model
        model = getModel(d['model'], d['secret'])    
        
        # Perform one-pass summarization
        timeline_df = generate_timeline(d['topic'],
                                        d['numevents'],
                                        model)
        #qas = pd.read_csv("results_tesla.csv")

        timeline_df.head(10)
        
        # Generate and return the HTML table with the summaries
        return generate_timeline_html(timeline_df)

    except BaseException as e:
        print(traceback.format_exc())
        return str(e)

setup_hf_backup(app)

# Start the FastAPI server
serve()